The Hall of Fallacies

Accent Fallacy

also known as: accentus, fallacy of accent

Taking something someone said and leaning on a different word, or cutting the quote short, to make it sound like they said something else.

In the group chat, Dana texts: "I don't have a problem with Sam's plan, I have a problem with the deadline attached to it." A coworker screenshots only the first half and reads it back out loud in a flat, confident voice: "She said she doesn't have a problem with Sam's plan," stopping right before the comma and stressing it like a settled fact. Everyone in the room now thinks Dana signed off on the plan itself, when what she actually said was closer to the opposite.

The tell: Something feels off even though the exact words match what was said or written. You find yourself wanting to say "that's not how I said it" or "read the whole thing" — the words are technically real, but the tone or the cut-off point is doing work the original sentence never asked it to do.

The move: A statement S is uttered or written under one stress/emphasis/punctuation/quotation pattern with content C. The arguer restates S under a different stress, emphasis, or quotation boundary, producing apparent content C*, then treats C* as if it were what was actually said and draws a conclusion from C* rather than from C.

Why it fails: Stress, emphasis, punctuation, and quotation boundaries are load-bearing parts of meaning, not decoration. Changing them changes what proposition is on the table. The argument's conclusion is licensed only by C*, the re-accented or re-clipped version, but the premise on offer was actually C. Because C and C* are not the same claim, the inference silently substitutes a claim the original speaker never made for the one they did, which is an equivocation across two readings of one string of words rather than a move from evidence to conclusion.

• Would changing the vocal stress or typographic emphasis on this exact sentence change what it claims?
• Does the quoted or reported material preserve the original emphasis, punctuation, and surrounding context of the source, or has a boundary been moved?
• Is the shift in stress, emphasis, or clipping doing the actual argumentative work here, rather than anything the speaker stated?
• If you read the original at a flat, neutral emphasis with the full surrounding sentence restored, does the 'gotcha' meaning survive or disappear?

Legitimate when: Re-emphasizing or re-punctuating a statement is legitimate when it faithfully reproduces the stress or context the original speaker actually intended or supplied themselves (for example, quoting someone's own italics, their own scare quotes, or a contrast they clearly drew), so that C* just is C rather than a rival reading manufactured by the arguer.

Ancient: Aristotle names accent (in Greek, prosody, later rendered in Latin as accentus) among the fallacies that depend on language. His illustration turns on a Greek word that reads one way and means 'where' but, given a different accent, reads as 'not' — since accent marks were not customarily written down in his day, a disputant could exploit the gap between the written word and its intended pronunciation to shift its sense. — Aristotle, On Sophistical Refutations

Built from: Wikipedia · Stanford Encyclopedia of Philosophy — Fallacies · The Fallacy Files — On Accent · Logically Fallacious — Accent Fallacy

Accident Fallacy

also known as: accidens, fallacia accidentis

Someone takes a rule that's only usually true and rams it onto the one case that's obviously the exception.

Your roommate says, 'Any pot left in the sink overnight draws ants, so anything sitting there gets tossed.' Fair rule, most of the time. But then he dumps out the pot you deliberately left soaking overnight with the lid on to loosen burnt rice off the bottom — no food scraps exposed, sealed, nowhere near the trail the ants use. He's not wrong about the rule in general. He's wrong to slap it on the one pot that plainly doesn't fit it, and to act like pointing that out is you just making excuses.

The tell: You find yourself saying 'but that's not what happened here' or 'my situation is different,' and instead of engaging with why it's different, the other person just repeats the general rule again, louder, as though repeating it settles it.

The move: A general rule or rule of thumb R holds for the typical/normal members of a class ('Xs are, as a rule, Y'). Case c is a member of the class but has an accidental (i.e., non-essential, situation-specific) feature that places it outside the conditions under which R actually holds. R is nevertheless applied to c without regard to that disqualifying feature, and the arguer concludes that c is Y. In scholastic terms: a dicto simpliciter ad dictum secundum quid — reasoning from a statement asserted without qualification (simpliciter) to a particular case that in fact requires qualification (secundum quid).

Why it fails: The rule was never true unconditionally; it was true only 'as a general rule,' i.e. under an implicit ceteris-paribus clause covering ordinary, unexceptional instances. The argument silently drops that qualification and treats the generalization as if it applied without exception. It therefore transfers a predicate that belongs to the subject only in its normal/essential circumstances onto a case where the predicate attaches, if at all, only accidentally (or not at all) — conflating what is true 'of the thing' in general with what is true 'of the thing under its accidental circumstances.' The gap between the rule's stated scope and its real, tacitly-qualified scope is where the inference breaks.

• What general rule or rule of thumb is doing the work here, and is it being stated as if it holds absolutely, with no exceptions?
• What normal-case conditions does the rule quietly assume in order to be true at all?
• Does the specific case under discussion have some accidental, atypical feature that takes it outside those normal conditions?
• Would the reason the rule exists in the first place actually still be served by applying it here — or is that reason absent or reversed in this case?

Legitimate when: The move is legitimate whenever the case genuinely does fall within the rule's ordinary scope (no disqualifying exception actually applies), or when the rule really is meant and defensible as exceptionless rather than as a mere generalization — in which case applying it to any member of the class is not an accident fallacy at all, just a sound application of a true universal.

Ancient: Aristotle lists 'accident' (para to symbebekos) among the fallacies that do not depend on language. As the Stanford Encyclopedia of Philosophy's entry on fallacies glosses it, the error occurs when 'some attribute belongs similarly to the thing and to its accident' — that is, when what is true of a subject only through some incidental, non-essential circumstance is treated as though it were true of the subject as such, or vice versa. The stock illustration used to explain this in the secondary literature involves Coriscus and Socrates: Coriscus is other than Socrates; Socrates is a man; therefore Coriscus is other than a man — the argument illegitimately carries a fact true of Socrates only accidentally (that this particular man is called Socrates, distinct from Coriscus) over onto the general term 'man.' Because the exact textual location and wording of this specific illustration were not independently re-verified against a primary translation during this research pass, confidence is marked medium rather than high. — Aristotle, Sophistical Refutations (Peri Sophistikōn Elenchōn), ch. 5, 166b28ff

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Accident (fallacy) · Internet Encyclopedia of Philosophy — Fallacies (Fallacy of Accident) · The Fallacy Files — Accident

Ad Fidentia

Making someone doubt themselves and then treating that self-doubt as if it proved they were wrong.

Three friends are splitting a restaurant bill in a group chat. One types, "pretty sure the tip's already on there, the total looks right." Another friend fires back, "are you SURE though? Like 100% sure? Because if you're not totally sure you shouldn't be doing the math." She keeps needling until the first friend types "...I guess I'm not 100% sure" and Venmos everyone extra for tip. Nothing on the receipt changed. Nobody looked at the receipt again. The only thing that moved was how sure she felt.

The tell: Notice that nobody actually pointed to anything wrong with what was said -- they just kept pushing until you weren't sure of yourself anymore, and then acted like that settled it.

The move: S asserts claim P. Instead of addressing P's supporting reasons or evidence, the challenger presses S on how confident or certain S feels about P ('Are you sure?' 'Can you be 100% certain?'), causing S's confidence to waver. The challenger then treats S's wavering confidence itself as grounds for rejecting P, without ever having offered a reason to doubt P.

Why it fails: A claim's truth does not covary with the speaker's subjective certainty about it. Confidence is a psychological state of the arguer; it carries no evidential weight regarding whether P is true. Treating a drop in confidence as if it were a concession or a refutation substitutes an attack on the person's inner state for engagement with the reasons behind the claim -- the same relevance failure at the core of ad hominem, just aimed at certainty rather than character or motive.

• Is the challenge aimed at how sure the speaker feels, rather than at the evidence or reasoning behind the claim?
• Would the claim's truth value change at all if the speaker felt completely confident versus completely unsure?
• Has the challenger supplied any counter-evidence, or only a demand for total certainty?
• Is ordinary honesty about a claim being open to revision (a normal feature of most empirical claims) being mischaracterized as weakness or surrender?

Legitimate when: It is legitimate criticism, not this fallacy, when the claim under discussion is itself a claim of certainty -- e.g. someone asserts 'I am certain P' as their actual argument for P (appeal to their own confidence), or an eyewitness's reliability turns on how sure they really were. In those cases confidence is the thing being argued about, not a stand-in for it, so probing it addresses the real substance rather than diverting from it.

Built from: Logically Fallacious (Bo Bennett) - Ad Fidentia · Stanford Encyclopedia of Philosophy - Fallacies

Ad Hoc Rescue

When proven wrong, instead of admitting it, someone bolts on a brand-new excuse that just happens to explain away the bad news, and refuses to say what would ever change their mind.

Your coworker swears a new productivity app will get everyone through their to-do list by Friday. Friday comes and half the list is still open. 'That's because the app needs a two-week adjustment period,' he says. Two weeks later, still behind: 'It's because the team isn't using it right.' A month after that, still behind: 'Honestly, without the app things would be so much worse, you just can't see it.' Notice the pattern: every single time the app fails to deliver, a fresh explanation appears out of nowhere, built for that exact moment, and somehow the app is never the problem.

The tell: You notice the story keeps changing shape to dodge whatever just went wrong, and you realize that no matter what happened, they'd have found a way to say they were right all along.

The move: Thesis T predicts observation E. Not-E is observed instead. Rather than revising or abandoning T, the arguer introduces auxiliary hypothesis A — invented solely to reconcile T with not-E, motivated by nothing except the need to save T, and carrying no independent testable consequences of its own. T is then declared to stand, undamaged, as if not-E had never been observed.

Why it fails: The defect is not that saving a thesis with an auxiliary hypothesis is always illegitimate — under the Duhem-Quine thesis, any single observation underdetermines which piece of a web of belief (the core theory or a background assumption) must be revised, so some revision is always logically available. The fallacy occurs when the added hypothesis is purpose-built for this one embarrassment, has no independent motivation or evidence, and generates no new predictions that could themselves be checked. At that point the 'rescue' isn't a reasoned revision of the belief network, it's a patch that makes the thesis unfalsifiable in practice: any future disconfirming evidence can be met with another equally ad hoc patch, so the thesis is insulated from the world rather than answerable to it.

• Was the rescuing hypothesis introduced only after the disconfirming evidence appeared, with no independent motivation?
• Does the rescue generate any independently testable predictions of its own?
• Is there any conceivable evidence the arguer would accept as refuting the thesis?
• Could the same rescue move be run to save the opposite thesis just as easily, showing it does no real explanatory work?

Legitimate when: Revising an auxiliary assumption instead of the core thesis is legitimate reasoning, not a fallacy, when the new hypothesis is independently supported, makes its own testable predictions, and is not simply reissued each time a new disconfirmation shows up. Standard scientific practice routinely blames background assumptions (measurement error, an uncontrolled variable, an approximation) for an anomaly, and that is sound so long as the blamed assumption can itself be checked and could, in principle, turn out false.

Built from: Ad Hoc Rescue - Logically Fallacious · Wikipedia - Ad hoc hypothesis · Wikipedia - Duhem-Quine thesis

Ad Hominem (Abusive)

also known as: argumentum ad hominem, abusive ad hominem

Instead of answering what you said, someone insults who you are and acts like that settles it.

In the group chat, Maria suggests splitting the dinner bill evenly since everyone ordered about the same. Dev replies, 'Ha, of course the person who showed up 20 minutes late thinks that's fair. Nobody's listening to you.' Nobody in the chat actually explains why splitting the bill evenly is unfair — Dev just makes fun of Maria being late and the group moves on as if her idea got shot down.

The tell: You notice the conversation never actually touched the idea itself — it went straight from 'here's my point' to 'here's what's wrong with you' — and somehow the point got treated as dead anyway, even though nobody laid a finger on it.

The move: Speaker S asserts claim P. S has unfavorable personal trait, history, or attribute T (unrelated to the truth-bearing content of P). Therefore, P is false / need not be considered.

Why it fails: T, whatever it is, is not evidence about the world that P describes. P's truth depends on how things stand outside of S — on facts, reasons, or data — not on S's character, habits, appearance, or past conduct. Swapping an evaluation of that external evidence for an evaluation of the speaker commits a relevance failure: the premise (the insult) and the conclusion (P is false) are not logically connected, even though the abuse can be psychologically effective at making an audience stop listening. The fallacy is compounded because the attack usually supplies no independent reason to doubt P at all — it substitutes for argument rather than supplementing it.

• State T (the trait attacked) and P (the claim dismissed) separately — is there any stated bridge connecting T to the truth of P, or does the argument jump straight from insult to rejection?
• Would the same P, uttered by a speaker without T, be accepted or at least taken seriously? If yes, T is doing all the rejecting and the argument is not about P.
• Has any independent objection to the claim itself been offered, or does the insult carry the whole weight of the 'refutation'?
• Is T actually about S's reliability as a witness/testifier (credibility) or about S's fitness to be trusted with a task (practical trust) — in which case it may be relevant to a different, narrower question than 'is P true'?

Legitimate when: Personal facts about a speaker are legitimately relevant when the question at issue is genuinely about the speaker — their credibility as a witness, their reliability as a testifier where independent verification is unavailable, a conflict of interest bearing on a recommendation, or their fitness for a role of trust. In a courtroom, impeaching a witness's character for truthfulness is procedurally proper because the jury must weigh testimony, not a free-standing factual claim submitted for its own evaluation. What is never legitimate is treating T as if it settled the truth of a separately verifiable proposition P.

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Ad hominem · Logically Fallacious — Ad Hominem (Abusive)

Ad Hominem (Circumstantial)

also known as: argumentum ad hominem, circumstantial ad hominem

Someone says you only believe what you're saying because it benefits you, and acts like that proves you're wrong.

At a family dinner, your uncle who works at a solar panel company mentions that solar is now cheaper than the electric company's rates in your area. Your cousin rolls her eyes and says, 'Of course you'd say that, you sell the things.' Nobody looks up the actual electric bill numbers or asks your uncle where he got his figures — the table just decides he's wrong because he works there.

The tell: You realize nobody actually checked the numbers or the facts. The whole objection was about who was talking, not about whether what they said holds up.

The move: Speaker S asserts claim P. S is in circumstance C (a job, a payment, a membership, a relationship, a position to gain or lose something) such that S would benefit from P being believed. Therefore, P is false / need not be considered.

Why it fails: C is a fact about S's situation, not a fact about the world P describes. Motive can explain why someone would say P; it cannot by itself show that what they said is untrue. The mere possibility that S's interest colored S's judgment does not establish that it did, and even if it did, P's truth is settled by the evidence for P, not by who benefits from P. The argument substitutes a psychological hypothesis about the speaker (why they talk this way) for an epistemic assessment of the claim (whether things are this way), so premise and conclusion are not logically connected — a self-interested person can still be reporting the facts correctly, and a disinterested one can still be wrong.

• Does the alleged interest actually change the evidence being offered, or does it only supply a motive for offering it?
• Is independent corroboration available — could the claim be checked against a source with no stake in the outcome — and if so, has anyone actually gone and checked it, or was the interest used as a substitute for checking?
• Is the interest demonstrated (a documented payment, a stated stake, a formal role) or merely conjectured from the speaker's job, affiliation, or group membership?
• Would the same evidence, presented by someone with nothing to gain, be accepted? If yes, the interest is doing all the rejecting, not the evidence.

Legitimate when: A speaker's circumstances are legitimately relevant to how much independent weight their unverified word should carry — this is ordinary witness-bias impeachment: a jury may discount interested testimony without deciding it is false, and a reader may reasonably ask for corroboration before taking a salesperson's or a lobbyist's word alone. What is never legitimate is treating the interest as if it had refuted the claim, or as license to ignore evidence that has already been produced and can be checked on its own terms.

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Ad hominem · Internet Encyclopedia of Philosophy — Fallacies · Logically Fallacious — Ad Hominem (Circumstantial)

Ad Hominem (Guilt by Association)

also known as: argumentum ad hominem, guilt by association

Making you dislike an idea just by pointing out that someone unlikeable also likes it — even though that has nothing to do with whether the idea is actually good.

In the group chat, someone shares a great budgeting app they've been using. Marcus replies, 'lol that's the app my ex used, and she was an absolute nightmare with money — hard pass.' Nobody has said one word about whether the app actually helps people save money. Marcus just took his leftover bad feelings about his ex and pasted them onto an app she happened to use too.

The tell: You notice the conversation quietly shifted from 'does this thing work' to 'who else likes this thing' — and somehow that swap felt like it settled something. A good gut check: imagine the exact same idea coming from someone likable instead. If the objection disappears, it was never about the idea.

The move: Person or claim X is linked to person/group G (X endorses G, is endorsed by G, resembles G, shares a category with G, or is merely near G). G has an unfavorable property or reputation. Therefore, X — or X's specific claim — is discredited, without any further premise showing that the link actually transmits G's fault to X or to X's claim.

Why it fails: The argument treats shared membership in a group, or mere proximity to a disliked party, as if it were evidence about the truth or merit of a separate, specific claim. But an association is not automatically a channel through which a trait travels: it can be incidental, involuntary, partial, or entirely unrelated to the content of the claim at issue. Even when the association is real and voluntary, the disliked party's endorsement adds no new evidence about the world the claim describes — a true claim stays true and a bad plan stays bad regardless of who else likes it. The fallacy substitutes an evaluation of the company a claim keeps for an evaluation of the claim itself, which is a relevance failure: the premise (the association) and the conclusion (the claim is false or the person is bad) are not logically connected, only emotionally connected via transferred disapproval.

• What exactly is the association, and is it voluntary, substantive, and directly about the claim at issue — or is it incidental, one-sided, coincidental, or several steps removed?
• Does the disliked party's endorsement or involvement change any actual evidence bearing on the claim, or does it only change how the audience feels about the claim?
• Would the identical claim, made or held by a neutral or admired party instead, be treated as equally suspect? If not, the association — not the claim's merits — is doing all the work.
• Is the 'guilt' being assigned a property of the claim itself, or only a property of some of the people who happen to hold it?

Legitimate when: The association is legitimate evidence when it reveals a genuine, direct, and relevant link between the disliked party and the specific claim — for example, the person actively adopted and now promotes the group's particular reasoning, took material support from them in exchange for advocacy, or the shared trait is precisely the one in dispute (e.g., a group's demonstrated pattern of fraud is relevant when evaluating whether to trust its new financial pitch). In such cases the move shades into a legitimate circumstantial ad hominem — it gives a reason to scrutinize possible bias or motive, but even then it only justifies added scrutiny, not outright dismissal of the claim on its own terms.

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Association fallacy · Internet Encyclopedia of Philosophy — Fallacies · Logically Fallacious — Ad Hominem (Guilt by Association) · The Fallacy Files — Guilt by Association

Ad Hominem (Tu quoque)

also known as: argumentum ad hominem, tu quoque, you too

Dodging criticism by pointing out the critic is a hypocrite, instead of answering what they actually said.

In the group chat, Priya posts: 'We really should stop ordering so much takeout, it's wrecking the budget.' Marcus fires back: 'You ordered pad thai literally last night, so I'm going to go ahead and ignore this.' Nobody in the chat actually addresses whether cutting takeout would help the budget — Marcus just made Priya's hypocrisy the whole topic, and the group moves on like the budget point got answered.

The tell: Notice that the actual point never got touched. Someone brought up a fact about the *person* talking instead of the *thing* they said, and somehow that felt like enough to end the conversation.

The move: Speaker S asserts claim X (often a norm, warning, or recommendation). Listener L observes that S's own conduct is inconsistent with X — S does not practice what X calls for. L concludes: therefore X is false, or S's case for X may be dismissed.

Why it fails: The truth of X depends on the world, not on the biography of whoever states it. A hypocrite's inconsistency is evidence about the hypocrite's character or sincerity, but it supplies no evidence bearing on whether X itself is true. Treating 'S fails to live up to X' as if it settles 'X is false' swaps a question about the speaker for the question actually on the table, so the criticism of X stands or falls untouched by anything just said about S.

• Would the original claim remain exactly as true or false if a perfectly consistent person had said it instead?
• Is the alleged inconsistency being used to deny the claim's truth, or only to question the speaker's standing or sincerity in pressing it?
• Does the speaker's inconsistent conduct actually appear as a premise anywhere in the argument being dismissed, or is it entirely external to it?
• If the inconsistency were resolved tomorrow (the speaker reformed), would that change whether the original claim is true?

Legitimate when: Pointing out inconsistency is not always fallacious. It is legitimate when the inconsistency bears on the speaker's sincerity, credibility as a witness, or standing to demand compliance from others (compare the equitable 'clean hands' doctrine, which withholds a remedy from an inconsistent party without pretending the underlying legal rule is false). It is also legitimate, following Locke's original sense, when used to expose that an opponent's own stated principles already commit them to a conclusion they are resisting — that is argument from the opponent's premises, not a claim that those premises are false because of who holds them.

Ancient: Jesus anticipates that his hearers will throw the proverb 'Physician, heal thyself' back at him — deflecting his message by pointing to his own condition rather than answering what he actually said. It is one of the oldest widely cited instances of the pattern, though the proverb is not presented there as a named fallacy, and the passage's original point was about expected miracles rather than a formal logic lesson. — Gospel of Luke 4:23 (quoting an older proverb)

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Tu quoque · Internet Encyclopedia of Philosophy — Fallacies

Affirmative Conclusion from a Negative Premise

Two things being ruled OUT of a group doesn't add up to proof that something is ruled IN.

At a bake sale table, someone says: "None of the nut-free desserts are the chocolate cake. None of the vegan desserts are the chocolate cake. So the chocolate cake must be the gluten-free one." Both starting facts are just saying what the cake ISN'T — not nut-free, not vegan. Neither one says a single word about gluten. Yet the speaker jumps straight to a confident, positive claim about gluten anyway, as if two "it's not that" facts had quietly turned into a "here's what it is" fact.

The tell: You notice every piece of evidence offered was really a story about what got crossed off a list — and then, out of nowhere, the final claim names a winner. The leap from crossing-off to naming feels like it skipped a step, because it did.

The move: A categorical syllogism in which at least one premise is negative (an E or O proposition, asserting that a class is wholly or partly excluded from another) but the conclusion is affirmative (an A or I proposition, asserting inclusion). Schematically: No M is P (or Some M is not P); [some relation between S and M]; ∴ All S is P (or Some S is P).

Why it fails: A negative premise asserts that two classes are, wholly or partly, separated from one another; it supplies no information about positive overlap. An affirmative conclusion asserts that two classes do, wholly or partly, coincide. No amount of separation-information can by itself certify a coincidence-claim — the two kinds of claim point in opposite directions. This is codified as one of the classical rules of the valid categorical syllogism: any syllogism with a negative premise must have a negative conclusion (and conversely, a negative conclusion requires a negative premise). Violating it produces a syllogism that is invalid by structure alone, regardless of whether the individual premises happen to be true; a Venn diagram in which the premises are satisfied and the conclusion fails can always be constructed.

• Translate each premise into standard categorical form (A: All S is P / E: No S is P / I: Some S is P / O: Some S is not P) — is at least one premise an E or O proposition?
• Is the conclusion an A or I proposition (affirmative) despite that negative premise?
• Can you build a model or Venn diagram where the negative premise(s) hold true but the classes named in the conclusion do not actually overlap?
• Strip the argument to its bare exclusion claims — does anything in them actually say two things belong together, or only that something is kept apart from something else?

Legitimate when: There is no legitimate version of this exact move: in standard categorical logic a negative premise never licenses an affirmative conclusion. The syllogism is repairable only by changing the conclusion to a matching negative one (or by discarding/rewriting the negative premise), not by any special case that rescues the affirmative claim.

Ancient: Aristotle's survey of the syllogistic figures establishes which premise combinations yield no valid conclusion at all, including the results underlying the later rule that nothing affirmative can be concluded when a premise is negative; the explicit rule itself is a Scholastic-era codification of Aristotle's results rather than a sentence found verbatim in his text. — Aristotle, Prior Analytics, Book I

Built from: Wikipedia · The Fallacy Files

Affirming a Disjunct

Somebody says 'it's this or that,' you confirm one of them is true, and they act like that proves the other one is false — but 'or' usually doesn't work that way.

In the group chat, someone writes: 'Either Priya's stuck in traffic or she's still on her call with the client — that's why she's late.' A minute later her calendar shows the client call is indeed still running. 'Oh good,' the person replies, 'so she's not stuck in traffic then, she'll be here any minute.' But of course she could easily be doing the call from her car while stuck in traffic — confirming one reason for being late did nothing to rule out the other.

The tell: That nagging 'wait, couldn't both of those be true at the same time?' feeling — nobody ever showed the two things were impossible together, they just acted like proving one killed the other.

The move: P ∨ Q (P or Q, where 'or' is the ordinary inclusive disjunction); P; ∴ ¬Q. (The mirror-image form, from Q instead, licenses ¬P instead.)

Why it fails: Ordinary logical disjunction is inclusive: 'P or Q' is true whenever at least one of P, Q is true, including the case where both are true. Confirming P therefore tells you nothing about Q's truth value — P and Q were never asserted to be mutually exclusive by the disjunction itself. The inference silently swaps in exclusive-or (XOR), under which affirming one disjunct really would rule out the other, for the inclusive-or that the premise actually expressed. That swap is an equivocation on the meaning of 'or,' and it is what makes the argument formally invalid rather than merely unpersuasive: there exist truth-value assignments (P true, Q true) that make every premise true and the conclusion false.

• Was the 'or' in the premise ever established as exclusive, or is exclusivity being smuggled in only by the shape of the argument that follows?
• Can you construct a case where both disjuncts are true at once? If so (the Wikipedia example: 'Max is a mammal or Max is a cat; Max is a mammal' — Max can be, and is, both), the inference is already broken.
• Do the two alternatives exclude each other as a matter of logic or fact, or only by rhetorical framing ('either you're with us or against us') that never argued for the exclusivity it assumes?
• Is this being confused with disjunctive syllogism? That form denies a disjunct to affirm the other (P ∨ Q; ¬P; ∴ Q) and is valid under inclusive-or — check which direction the argument actually runs.

Legitimate when: The inference is valid whenever the disjunction genuinely is exclusive — either by explicit stipulation ('exactly one of the following holds') or because the disjuncts are logically or physically incompatible (a number can't be both strictly less than 5 and strictly greater than 10 at once, so confirming one truly does rule out the other). Outside of a demonstrated exclusive-or, the move is unsound.

Ancient: Diogenes Laertius records the Stoic fifth indemonstrable: from a disjunction and one of its disjuncts, infer the negation of the other. For the Stoics this was a valid move because their disjunction was built to be exclusive; the fallacy as understood today arises only when the identical inference is drawn from an ordinary inclusive 'or.' — Diogenes Laertius, Lives of Eminent Philosophers, Book 7 (Stoic logic)

Built from: Wikipedia — Affirming a disjunct · The Fallacy Files — Affirming a Disjunct

Affirming the Consequent

Just because the thing you expected to see showed up doesn't mean your favorite explanation is what caused it.

In the group chat, Mia types: "If Jordan won the raffle, he'd be posting a victory selfie right about now. He just posted a victory selfie. So Jordan won the raffle!" Except Jordan posts a victory selfie practically every other day — for free pizza, for a good parking spot, for finishing a crossword before his coffee got cold. The selfie showing up doesn't prove anything about the raffle specifically; it just proves Jordan likes posting selfies.

The tell: You get a nagging feeling that the explanation clicked a little too fast — like one single clue got used to lock in one single story, when that exact same clue would fit a dozen other stories just as comfortably.

The move: If P then Q; Q is true; therefore P is true. (P → Q, Q, ∴ P)

Why it fails: The conditional "if P then Q" only asserts that P is sufficient to produce Q — it says nothing about whether P is the only thing that can produce Q. Affirming Q and concluding P treats the one-way conditional as though it were reversible (as though P ↔ Q), which the original statement never established. Q can be true for reasons that have nothing to do with P, so its truth gives no reliable evidence that P occurred.

• Does the truth of Q actually rule out every explanation besides P, or does it just fail to rule any of them out?
• Could Q be true even if P were false — is there a rival cause or condition that produces Q just as well?
• Was the original claim really "P guarantees Q," or has it quietly been treated as "only P produces Q" (a hidden if-and-only-if that was never argued for)?
• List the other conditions that could make Q true — have they actually been investigated and ruled out?

Legitimate when: The move is legitimate only when the converse (Q → P) has been independently established — i.e., when P and Q are genuinely equivalent (P if and only if Q), so that Q really does uniquely point back to P and nothing else could have produced it.

Ancient: Aristotle observes that people slide from "if A is true, B is necessarily true" into wrongly believing "if B is true, A is necessarily true" as well. His example: because a person with a fever is hot, people wrongly infer that anyone who is hot must have a fever — even though heat can just as easily come from exercise, sunshine, or a warm room, not only from fever. — Aristotle, On Sophistical Refutations, ch. 5 (169b19)

Built from: Wikipedia · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies

Alleged Certainty

Saying "everybody knows this" instead of actually showing why it's true.

In the group chat, someone types: "Obviously the new manager only got the job because she's friends with the boss -- everyone in the office says so." Nobody has looked at how the hiring actually went. The claim just rides in on the words "obviously" and "everyone says," and by the time anyone thinks to ask for an actual reason, the chat has already moved on to arguing about what to do about it.

The tell: You notice you never actually heard a reason -- you just heard confidence, and somehow that confidence made asking for a reason feel awkward, like you'd be the odd one out.

The move: Speaker asserts P while attaching certainty-markers ("everyone knows," "obviously," "it is simply a fact," "there's no question that...") in place of evidence for P. Therefore, P.

Why it fails: The certainty markers smuggle in a second, unsupported claim on top of P itself: that P is so settled a reasonable person could not doubt it. Neither claim is evidenced. The confident tone does the work a premise was supposed to do, and it pre-empts the burden of proof by declaring the question closed before any support is offered. Even when P turns out to be true, no argument for P has actually been made.

• What evidence has actually been offered besides the assertion that the claim is certain?
• Would an informed, good-faith person actually disagree with this claim -- and if so, why does the speaker talk as if they wouldn't?
• Strip out the certainty language ("obviously," "everyone knows," "simply a fact") -- what support is left standing?
• Is this genuinely settled background knowledge, or is it actually the very point that's in dispute?

Legitimate when: The claim really is uncontested common ground between the parties (e.g., "without water, we cannot survive") such that stating it plainly wastes no one's time, or it is legitimate shorthand for a well-established expert consensus the speaker could substantiate on request. The move fails precisely when the 'certainty' is doing the persuading instead of standing in for evidence that exists but is merely omitted.

Built from: Logically Fallacious (Bo Bennett) · Stanford Encyclopedia of Philosophy - Fallacies · Wikipedia - Proof by assertion · Wikipedia - List of fallacies

Alphabet Soup

Burying a claim under so many acronyms and technical terms that people get confused and just assume the speaker must know what they're talking about.

At a school budget meeting, a consultant says: "Our approach leverages the MTSS Tier 2 framework aligned with ESSA subgroup metrics, cross-referenced against the district's IEP compliance data under the FERPA-safe reporting layer. So obviously the program is working." Nobody in the room actually knows what any of that means well enough to argue with it, so the room just nods. Later, a parent asks, "Okay, but what were last year's test scores, in plain numbers?" and there's an awkward pause, because the acronyms were standing in for a number that was never actually given.

The tell: You realize you nodded along or felt impressed, but if someone asked you to explain what was just said in your own words, you couldn't do it, and you're not even sure there was anything there to explain.

The move: Speaker S makes claim C while embedding it in a dense cascade of acronyms, abbreviations, and insider jargon (A1, A2, A3, ...). The audience cannot parse the acronyms quickly enough to evaluate whether they actually support C. S's fluent handling of the jargon is then treated, implicitly, as if it were evidence for C itself, and the audience's confusion or intimidation is treated as a reason to defer rather than as a reason to ask for the acronyms to be unpacked.

Why it fails: Fluency with a vocabulary is not the same thing as having evidence for a claim; a person can rattle off CGI, WYSIWYG, and PHP, or ACA, MLR, and QHP-SBM, with total accuracy and still have offered zero support for the conclusion on the table. The jargon cascade substitutes a performance of belonging to an in-group for an actual chain from evidence to claim, and it works specifically because the audience's inability to follow the shorthand is misread as the audience's inability to find fault with the reasoning, when in fact there may be no reasoning present to fault at all.

• Can the claim be restated in plain language without losing anything, and does it still seem supported once restated?
• Is each acronym or technical term doing real content work here, or could it be dropped without changing what's actually being claimed?
• When you unpack the jargon term by term, does any actual evidence, number, or mechanism for the conclusion show up, or just more labels?
• Would this same density of shorthand be normal, unremarkable shop talk between two people who already share the field, or is it being aimed at someone outside the field?

Legitimate when: Dense acronym and jargon use is legitimate, and not this move at all, when it is efficient shorthand between people who already share the field and could unpack it on request, and when the underlying claim is also separately supported by evidence the jargon is merely labeling rather than replacing.

Built from: Logically Fallacious (Bo Bennett) · Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — List of fallacies

Alternative Advance

It's when someone hands you two choices that both secretly agree to the same hidden yes, so no matter which one you pick, they win the real question you were never actually asked.

In the team group chat someone posts: 'Pizza from Tony's or from Luigi's for lunch tomorrow?' Nobody ever agreed the team was doing pizza at all — but three people reply with a place, and now it's treated as settled that pizza is happening, when that was the one thing actually still open.

The tell: That queasy 'wait, when did we agree to this part?' feeling — you realize you were only ever asked the small downstream question, never the bigger one that was supposed to come first.

The move: A questioner offers exactly two (or more) options, A and B, both of which presuppose a single contested proposition P (e.g. 'that you will buy,' 'that you agree to X'). The respondent is invited to choose only between A and B; whichever is picked, that choice is then treated as if it had settled P — even though P was never independently established or agreed to, and no option corresponding to rejecting P was ever on the table.

Why it fails: A and B are not genuine alternatives on the point that actually matters; they are cosmetic variants of the same unproven commitment. By making the visible decision (A vs. B) look like the whole decision, the arguer skips past — and treats as already settled — the prior, separate question of whether P should be accepted at all. It is a species of question-begging worked through a menu instead of a single question: the real disjunction (P or not-P) is suppressed, leaving a fake disjunction (A or B) that structurally cannot register a 'no'.

• Do both offered options presuppose the same contested commitment?
• Is a third option — including refusing the question's framing entirely — available and arbitrarily excluded from the menu?
• Would restating the choice without the embedded presupposition ('do you want this at all?') change the respondent's answer?
• Whose purpose is served by narrowing the field to exactly these two options rather than asking the more basic question first?

Legitimate when: When the underlying commitment P is not actually in dispute — already agreed to, established, or freely granted by the respondent — narrowing the remaining choice to sub-options of P is ordinary efficient dialogue, not a fallacy (e.g. a diner who has already ordered dessert being asked 'cake or pie?'). The move only fails when P itself is the thing still up for grabs and the menu is used to paper over that fact.

Ancient: Aristotle discussed a fallacy in which several questions are bundled and treated as if they demand a single answer — a respondent forced to answer yes or no is thereby made to concede a presupposition (such as prior wrongdoing) that was never actually established. Alternative Advance is a close descendant of this pattern: instead of one loaded question, two loaded options are offered side by side, but the trick is the same — any direct answer concedes ground that was never argued for. — Aristotle, Sophistical Refutations (the fallacy of 'many questions', plurium interrogationum)

Built from: Logically Fallacious — Alternative Advance · Wikipedia — Loaded question · Wikipedia — Complex question · Stanford Encyclopedia of Philosophy — Fallacies

Amazing Familiarity

Someone states, as flat fact, something they could not possibly know — like what a stranger was secretly thinking — and then builds their point on top of it.

In a group chat about a coworker who got a promotion, someone writes: 'Honestly, the second she got the offer she was laughing at all of us behind our backs, thinking we're idiots for staying so loyal. That's exactly why she doesn't deserve it.' Nobody in the chat was in the room. Nobody has a text or quote. The writer just declares what was going on inside another person's head as if it were a fact they witnessed, then uses that invented inner life to justify the verdict.

The tell: A flash of 'wait, how could you possibly know that?' — the claim is about someone's private thoughts, an event nobody watched, or something nobody ever recorded, yet it's delivered with the flat confidence of someone reading off a receipt.

The move: Speaker asserts fact F about a matter that is private, unwitnessed, or in-principle inaccessible to any outside observer (another person's unspoken thoughts, an unrecorded event, the exact preferences of a deity, etc.). F is stated flatly, with no cited source and no hedge, as though the speaker had direct access to it. A further claim or conclusion C is then built on F as though F were established.

Why it fails: The premise smuggles in an epistemic vantage point the speaker could not actually occupy. Nothing is offered — no testimony, record, instrument, or inference chain — that could get an ordinary observer from 'outside the fact' to 'knowing the fact.' The argument therefore trades on a claim that is, for this speaker in this situation, unverifiable in principle, not merely unverified in practice; once F is exposed as asserted-not-known, whatever rides on it (C) loses its support. It differs from a merely unsupported premise because the very nature of the claim (private mental states, unwitnessed history, divine intent) rules out ordinary evidence ever backing it for this speaker.

• Could the speaker actually be in a position to know this fact — through observation, testimony, or record — or is the fact by its nature private/unwitnessed/unrecorded?
• What source is being claimed for the knowledge, and is that source named, checkable, or even accessible in principle?
• Is the claim stated with the certainty of an eyewitness report, when at best it could only be an inference or a guess?
• If the unknowable claim is bracketed out or downgraded to 'I suspect,' does the conclusion still stand on its own?

Legitimate when: The claim is accompanied by an actual epistemic route to it — direct testimony, documents, recordings, expert inference from evidence, or an explicit hedge ('I believe,' 'it appears,' 'the pattern suggests') that marks it as a probabilistic judgment rather than a report of fact. Confident claims about another's mental state are also legitimate when drawn from that person's own stated words or well-established behavioral patterns, rather than asserted as bare, unsourced fact.

Built from: Logically Fallacious (Bo Bennett) — Amazing Familiarity · Logically Fallacious (Bo Bennett) — Amazing Familiarity (tool page) · Wikipedia — List of fallacies (checked; no dedicated entry found for this term)

Ambiguity Fallacy

Someone uses one word to mean two different things in the same argument, and slides from the harmless meaning to the loaded one without telling you.

At the family dinner table, your uncle says: 'Everyone deserves a seat at the table — that's basic decency. My neighbor put up a fence so people can't cut through his yard anymore. So he thinks he's better than everyone, denying people their seat at the table.' 'A seat at the table' started out meaning something like being heard and respected in general. By the end it's been quietly swapped for a literal claim about a physical yard, as if refusing trespassers were the same thing as denying someone dignity. Nobody agreed to that second meaning — it just got smuggled in under the first one's good reputation.

The tell: You nod along at the start because the phrase sounded true and reasonable, then feel a jolt of 'wait, that's not what that word meant a second ago' once the same phrase gets used to nail down a much bigger, different point.

The move: An expression, phrase, or construction E occurs (or has a counterpart) at two or more points in an argument. On one occurrence E carries reading M1; on another it carries reading M2 (a different word sense, a different grammatical parse, a different scope, or a different referent). The argument's steps are only licensed if E keeps one fixed reading throughout, so the move from premises to conclusion silently trades on the switch from M1 to M2. Restated with a single reading fixed for E, the inference collapses.

Why it fails: Validity requires that a repeated term or construction mean the same thing everywhere it does logical work. When E shifts sense, the argument is really trading a claim that is true-under-M1 for a claim that is only interesting-or-wanted under M2 (or vice versa); no single, consistent statement was ever established that supports the conclusion. This is exactly the classical fallacy of four terms in a syllogism: a middle term used in two senses functions as two different terms, so the syllogism only has the surface form of three terms while actually needing four to go through, and no valid three-term syllogism exists to license the conclusion.

• Which specific word, phrase, or grammatical construction changes its reading between one premise and another (or between a premise and the conclusion)?
• If that expression is replaced everywhere by an explicit, single, fixed definition (or an unambiguous rewrite), does the inference from premises to conclusion still go through?
• Is the ambiguity lexical (one word, two dictionary senses), structural (one sentence, two grammatical parses), or a shift in scope/reference (e.g., 'right' as moral ideal vs. legal entitlement, 'end' as termination vs. purpose)?
• Would a listener who is only handed the second (shifted) sense still find the earlier premises persuasive, or does the persuasive force secretly depend on the audience carrying over the first sense?

Legitimate when: The repeated or ambiguous-looking expression is in fact used with exactly one meaning and one grammatical structure throughout, and any apparent double meaning is resolved by context, explicit stipulation, or disambiguating rewrite before the inference is drawn; deliberate wordplay, pun, or figurative language used for stylistic effect (not to smuggle in a false inferential step) is likewise not the fallacy.

Ancient: Aristotle catalogues refutations that depend on language (para tēn lexin) as a distinct class from refutations that depend on the facts, and within that class distinguishes homonymy — a single word or name carrying more than one meaning — from amphiboly, ambiguity arising from a sentence's grammatical structure rather than any one word. Both, he argues, produce arguments that look conclusive only because a term or phrase is silently allowed to shift its sense between occurrences. — Aristotle, On Sophistical Refutations

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Equivocation · Wikipedia — List of fallacies

Anonymous Authority

Someone backs up a claim by pointing to "experts" or "sources" they refuse to name, so you can't actually check who said it.

In the family group chat, your cousin sends: 'Don't buy that brand of baby formula, by the way — I heard from someone who works at the company that they cut corners on testing.' Everyone stops buying it immediately. When you ask who told her, she says, 'I promised I wouldn't say, but trust me, it's a real person and they know what they're talking about.'

The tell: You notice there's no name, no article, no document to click on — just a person and a claim you're supposed to take on faith, dressed up to sound like it came from someone who'd know.

The move: An unnamed or unidentifiable source S* ("experts say," "studies show," "sources close to the matter," "they say") is reported to hold that P. Therefore, P.

Why it fails: An appeal to authority is only as strong as the bridge conditions Logically Fallacious lists for the parent fallacy: the source must actually be an authority in the relevant subject, must be trustworthy, must not be an outlier among disagreeing authorities, and must be quoted accurately rather than misquoted. When the source is left unnamed, every one of those conditions becomes uncheckable in principle, not just in practice — there is no name to look up a credential against, no way to confirm the person exists or said what is claimed, and no way to tell whether 'they' is one crank or a genuine consensus. The argument therefore doesn't just have a weak premise; it has a premise immunized against verification, so the appearance of expert backing does the persuasive work that actual, checkable expertise would otherwise have to do.

• Who exactly is the claimed authority, and can their identity and credentials actually be checked against something independent of this claim?
• Is there an innocent, checkable reason for the anonymity (a protected whistleblower, a cited-but-unlinked peer-reviewed study), or does the anonymity itself only serve to block scrutiny?
• Is any inspectable evidence offered besides the unnamed source's say-so — a document, a dataset, a named study one could look up?
• If a real name and credential were substituted for 'experts' or 'sources close to the matter,' would the claim's credibility survive, or does all the persuasive force come from the unverifiable label itself?

Legitimate when: When the anonymity is independently justified and does not block verification of the substance — e.g., a journalist protecting a whistleblower's safety while other named, checkable evidence corroborates the claim — or when the phrasing merely compresses a well-established, easily verifiable consensus that any reader could confirm elsewhere (Logically Fallacious's own example: 'scientists say the earth is in a warming phase' stands in for a checkable body of published climate research, not for a single unverifiable insider).

Built from: Logically Fallacious — Anonymous Authority · Internet Encyclopedia of Philosophy — Fallacies (Appeal to Authority) · Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Argument from authority

Anthropomorphism

Anthropomorphism is treating a thing that isn't a person as if it has feelings and motives, and then arguing based on those made-up feelings.

In the family group chat someone writes: 'The car is being stubborn again, it doesn't want to start because it's still mad I left it out in the rain all week. Honestly I think it's holding a grudge, we should just get rid of it.' Everyone nods along and starts shopping for a new car, and nobody pops the hood to check whether the battery terminals are corroded.

The tell: You notice the conversation has quietly stopped being about wires, fluids, and parts and started being about somebody's mood, grudge, or attitude, and a real decision is about to get made on that basis.

The move: Non-human entity X exhibits behavior or property B. A human exhibiting B would possess human mental state or trait M (intention, emotion, purpose, will). Therefore X possesses M. Conclusion C, which depends on X's possessing M, is asserted.

Why it fails: The move from an outward behavioral resemblance to a shared inner mental state is an unlicensed analogy: resembling human conduct does not establish that the entity has the human-specific psychological architecture (intentions, feelings, moral standing) the conclusion needs. The human vocabulary is imported for its emotional or explanatory pull while the actual mechanism producing B is left unexamined, so the argument trades on a metaphor as if it were a literal, evidenced fact.

• What human trait is being ascribed to the non-human subject, and on what evidence beyond the resemblance itself?
• Does the conclusion depend on reading the ascription literally, rather than as a dispensable metaphor or figure of speech?
• Would the argument survive being restated entirely without agential language (no wants, hates, punishes, decides)?
• Is the entity even the kind of thing that could, in principle, have the interior state being attributed to it?

Legitimate when: The human-shaped description is used self-consciously as a predictive or explanatory model rather than a literal claim — e.g. Dennett's 'intentional stance,' saying a chess program 'wants' to protect its queen, or a biologist saying a gene 'behaves selfishly' — as long as no conclusion is drawn that actually requires the entity to possess a human mind, and the speaker would readily unpack the phrase into non-agential terms if pressed.

Ancient: Xenophanes observed that mortals imagine the gods as born, and as wearing clothes, speaking, and having bodies just as they themselves do; and that if horses, oxen, or lions had hands and could draw or make art as humans do, horses would draw the gods as horses and oxen as oxen, each in its own likeness. He concluded that the divine, whatever it is, resembles humans neither in body nor in mind. — Xenophanes, fragments (preserved via later doxographers; standard attribution)

Built from: Wikipedia — Anthropomorphism · Logically Fallacious — Anthropomorphism · Internet Encyclopedia of Philosophy — Fallacies

Appeal to Accomplishment

Someone points to their trophy case instead of answering the actual question, or tells you your empty trophy case means you're not allowed to ask it.

In the group chat about whether the new team lead's rollout plan will actually work, someone posts: 'Bro has shipped four apps that hit a million downloads. Just trust the plan.' Nobody has looked at the plan. When a coworker raises a specific worry about the timeline, the reply is: 'Cool story, how many apps have you shipped though?'

The tell: You notice the conversation drifted from 'is this plan good' to 'is this person impressive,' and any pushback gets met with a resume instead of an answer.

The move: S has notable accomplishments A (in some domain, possibly unrelated to P); S asserts P; therefore P is true. Converse form: critic C lacks accomplishments comparable to S's; therefore C's objection to P fails.

Why it fails: Accomplishment is treated as a stand-in for the actual evidence or argument for P, but a track record of success — even a genuine one — does not by itself raise or lower the probability that any particular claim the accomplished person makes is true, unless the accomplishments were earned by developing exactly the competence the claim depends on. In the converse form, the defect is sharper still: a critic's lack of comparable achievement says nothing about whether the specific objection raised is correct, since evaluating an argument does not require having personally matched the feat being criticized.

• Are the cited accomplishments in the same domain as the claim, and do they plausibly bear on its truth?
• Is any evidence for the claim offered besides the speaker's resume or track record?
• Is a critic's lack of comparable accomplishment being used to dismiss the content of their objection without addressing it?
• If a less-accomplished person made the identical claim with the identical evidence, would it be treated as equally credible?

Legitimate when: The accomplishment is direct, relevant evidence of the specific competence the claim requires (e.g., a structural engineer who has designed dozens of safe bridges citing that experience for a claim about bridge-load tolerances), and it is offered alongside — not instead of — the substantive case, with all parties agreeing the track record is a reliable proxy for expertise in this exact context.

Built from: Wikipedia — Appeal to accomplishment · Logically Fallacious — Appeal to Accomplishment · Stanford Encyclopedia of Philosophy — Fallacies (ad verecundiam / appeal to authority)

Appeal to Anger

Getting you mad on purpose so you'll say yes before you've had a chance to think it through.

In the group chat for a shared apartment, someone posts a photo of dishes left in the sink for the third time this week: 'This is SO disrespectful, right? Can you believe this?? Anyway, that's why I think we should just cancel movie night — nobody in this house cares about anyone else.' Everyone's fuming about the dishes within seconds. Nobody stops to ask what the dishes actually have to do with movie night.

The tell: You notice you're worked up, but if someone calmly asked you to walk through why the mad thing actually leads to the decision being pushed, you'd come up empty — the feeling arrived, but the reasoning never did.

The move: Anger is aroused in the audience toward some target (or the arguer displays their own anger), and this aroused anger is then offered as if it were the reason to accept conclusion P or adopt proposal A — with no premise actually linking the source of the anger to the truth of P or the merit of A.

Why it fails: Being angry is a state, not a reason. The anger may well be provoked by something genuinely wrongful, but the argument substitutes the intensity of that feeling for any fact connecting the wrong to the specific claim on the table. The heat of the reaction is used to carry the audience past the step where the grievance would actually have to be shown to bear on P or A.

• Is the anger aroused connected to any stated reason, or does it substitute for one?
• Would the claim or proposal still look supported if the audience read it calm, a day later?
• Is the wrong being described actually the matter under decision, or is it an adjacent grievance imported for heat?
• If you subtract the outrage-triggering language, is there anything left that speaks to the specific conclusion?

Legitimate when: When the very wrong that provokes the anger is the fact in dispute — for instance, victim-impact testimony offered to establish the severity of harm within a sentencing decision, or a person's angry reaction cited as evidence of how seriously they took a wrong, under a standard where genuine reactions are themselves probative. It stays legitimate only so long as the anger is tied to that specific finding rather than carried over to a different, merely adjacent proposal.

Ancient: Aristotle analyzes anger among the emotions a speaker can arouse, observing that people do not render the same judgments when angry as when calm, which is why orators work to put judges into the fitting emotional state. — Aristotle, Rhetoric, Book II

Built from: Stanford Encyclopedia of Philosophy - Fallacies · Wikipedia - Appeal to emotion · Wikipedia - List of fallacies · Logically Fallacious - Appeal to Anger

Appeal to Authority

also known as: argumentum ad verecundiam, appeal to modesty

Believing something just because an impressive-sounding person said it, without checking whether that person actually knows this particular thing.

In the family group chat, someone posts a screenshot: 'My dentist said this new detox tea flushes out toxins and boosts your immune system, and she's literally a doctor, so I'm ordering a case.' Nobody asks what a dentist knows about immune systems or detox drinks; the word 'doctor' does all the work, and three cousins immediately ask for the link.

The tell: You notice the conversation has stopped being about proof and started being about who said it — nobody can point to a study, a test, or a real result, they can only point to the person's title or fame, and somehow that feels like enough.

The move: Authority S asserts that P. S is (treated as) an authority. Therefore, P.

Why it fails: The move is only as good as the unstated bridge premises it skips: that S is actually an expert in the specific domain P falls under, that S is representing the real state of expert opinion rather than a fringe or superseded view, that S is not compromised by bias or conflict of interest, and that S's words are being accurately reported rather than distorted or stripped of qualification. When any of those conditions is missing, the argument substitutes the fact of an assertion for evidence for the assertion — it treats testimony about who believes P as if it were evidence for P itself. The Stanford Encyclopedia of Philosophy frames the core failure as accepting a pronouncement 'as evidence for a proposition' from someone 'taken to be an authority but [who] is either not really an authority or a relevant authority' — the defect is a relevance failure, not a rejection of expertise as such.

• Is S actually an expert in this exact, narrow domain — or an expert in something adjacent that merely sounds related (a Nobel physicist opining on economics, a doctor endorsing a supplement outside clinical trials)?
• Does S represent the broad, current weight of expert opinion in the field, or is S a lone dissenter, an outlier, or citing a view later experts overturned?
• Does S have a financial, institutional, or personal stake in P being believed, and is that conflict disclosed?
• Is S being quoted accurately and in context, with original hedges and caveats intact, or has a qualified claim been flattened into a bare assertion?

Legitimate when: Deferring to a genuine, domain-relevant, disinterested authority is a reasonable and often necessary shortcut when independently verifying every claim is impractical — this is defeasible reliance: it holds provisionally, is revisable in light of better evidence or a more qualified authority, and is strongest when the expert's claim falls within a field with real, checkable methods and a working consensus (e.g., citing a virologist on how a specific virus spreads).

Ancient: Cicero describes how Pythagoras's followers, when challenged to defend a claim in debate, would simply answer 'ipse dixit' — 'he himself said it' — letting the master's bare word close off further argument instead of offering reasons of their own. — Cicero, De Natura Deorum, Book I

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Argument from authority · Wikipedia — Ipse dixit

Appeal to Celebrity

Something must be true or good just because a famous person said it or is seen using it.

At the gym, a guy tells his workout partner, "Just get the pre-workout in the black tub — that huge bodybuilder with two million followers takes it every single day, and look at him." His partner orders a tub before reading a single ingredient on the label, because if it works for a guy built like that, it must work.

The tell: Notice you can explain WHO likes it, but not WHY it actually works — the reasons never show up, only the famous face.

The move: Person S is famous/admired (independent of any claimed expertise); S endorses, uses, or asserts P; therefore P is true, or product/cause X is good. (A fame-only special case of the appeal to authority: unlike the general form, no expertise is even alleged — mere recognizability is offered as the license.)

Why it fails: Fame tracks recognizability, not competence, and is frequently anti-correlated with relevant expertise (most celebrities endorsing a product are actors, athletes, or musicians with no training in it) and is often compromised by a paid or personal interest in the endorsement. Because the celebrity's involvement carries no evidential connection to the truth or quality of the claim, withdrawing the endorsement leaves the claim's actual support completely unchanged — the only thing doing rhetorical work is the audience's admiration for the person, not any reason bearing on the matter at hand.

• Does the celebrity have any domain-relevant expertise bearing on the specific claim, or only general fame?
• Is the endorsement being offered as evidence the claim is true, or merely as evidence the celebrity is popular?
• Would the claim's evidential standing change at all if the endorsement were withdrawn tomorrow?
• Is the celebrity paid or otherwise incentivized to endorse, and does that undercut the endorsement as disinterested testimony?

Legitimate when: When the celebrity's fame arises from, and coincides with, genuine domain expertise directly relevant to the claim (a renowned chef's opinion on knife quality, a Nobel laureate physicist's opinion on a physics question) — then the appeal functions as an ordinary, defeasible appeal to authority rather than a fame-only appeal to celebrity. It is also unproblematic when the claim being supported is merely autobiographical ("I use this and like it") rather than a general claim about the product's merit for others.

Ancient: Cicero recounts that when Pythagoras's followers were pressed to defend a claim, their standard answer was simply that the master himself had said so (the Greek autos epha, Latinized by Cicero as ipse dixit) — treating Pythagoras's revered name as sufficient proof and closing off further argument. Cicero criticizes the habit, insisting that the force of reasoning, not the authority of a reputation, is what should settle a dispute. This is not a modern celebrity endorsement, but it is a genuinely attested classical instance of the same underlying move that appeal to celebrity narrows and modernizes: letting a person's standing substitute for any reason bearing on the claim itself. — Cicero, De Natura Deorum

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Logically Fallacious — Appeal to Celebrity · The Fallacy Files — Appeal to Celebrity · Wikipedia — Argument from authority · Wikipedia — Ipse dixit

Appeal to Closure

Someone tells you to just accept an answer — not because it's been shown to be right, but because everyone is tired of asking the question.

A group chat has been arguing for weeks about whether a roommate secretly ate someone's leftovers. No one has any real proof either way. Finally one person writes: 'Look, we've been going in circles forever, this is exhausting everyone, let's just agree it was Jordan and never bring it up again.' Nobody actually found out anything new. The 'agreeing' only happened because the group was worn out, and once it's said out loud, asking about it again suddenly feels like the rude move — even though the actual question was never answered.

The tell: You notice the reason being given for accepting the answer is 'we're all so done with this,' not 'here's what we found out' — and that bringing the topic back up now feels like a social violation rather than a reasonable thing to do.

The move: A question Q has remained open for a long time. The people affected by Q want, and are entitled to feel, the relief that comes from Q being settled. Therefore, the currently favored answer to Q should be accepted as correct (or the currently favored action should be taken), and the case should be closed — not because the evidence for that answer has improved, but because settling is itself offered as the reason to settle.

Why it fails: The desire for closure is real and its psychological cost is real, but it is a fact about the audience's state of mind, not a fact about which answer to Q is true. The argument swaps an epistemic question ('does the evidence support this finding?') for a therapeutic one ('would accepting this finding make the pain stop?') and then answers the epistemic question with the therapeutic one, as if exhaustion were itself evidence. Nothing about how long an issue has been open, or how much people are hurting from its being open, changes the facts the issue is actually about — a report can be just as false after two years of waiting as it would have been on day one. The move also quietly forecloses future correction: because the payoff (relief) is collected the moment the matter is declared closed, the 'closing' clause does argumentative work that has nothing to do with whether the conclusion is right.

• Does the desire to move on bear on the truth of the finding being accepted, or only on how long people have had to wait for an answer?
• Who benefits from the matter being declared closed tonight, and is that benefit doing the work that evidence should be doing?
• Would this same conclusion be accepted on this same evidence if no deadline, exhaustion, or public pressure existed?
• Is the call to 'move on' paired with a promise never to revisit the question even if better evidence later appears?

Legitimate when: Wanting to stop investigating is not automatically fallacious. Legal and institutional systems build in deliberate stopping rules — statutes of limitations, res judicata, appeal deadlines, sunset clauses on inquiries — because unlimited re-litigation has its own real costs, and a society that never lets a matter rest cannot function. That is a legitimate procedural or practical judgment: given diminishing returns on further investigation, we choose to stop investigating and treat the current, evidence-based answer as final for practical purposes. The fallacy is specifically the further step of treating the desire for closure as evidence that the finding itself is correct. Ending an inquiry because further inquiry isn't worth its cost is reasonable procedure; declaring a finding true because ending the inquiry feels good is the fallacy.

Built from: Appeal to Closure - Logically Fallacious · Appeal to Closure - Logically Fallacious (tools/lp entry)

Appeal to Coincidence

Someone waves away a pattern that's staring everyone in the face by just saying "coincidence" — without ever checking how likely that pattern really is to happen by chance.

Every single time your roommate borrows your car, it comes back with an empty gas tank. Five times in a row now. You bring it up and they say, breezy and quick, "Huh, weird, must just be a coincidence," and grab their keys to leave before you can say anything else. They never actually explain why five-for-five is likely to happen randomly — they just use the word like a magic spell that ends the conversation.

The tell: The explanation arrives too fast, too smooth, and too final — nobody actually did any figuring, they just handed you a label and moved on, and the label happens to be the one that lets them off the hook.

The move: Evidence E (pattern, timing, mechanism, repeated correlation) supports X caused/explains Y. Arguer: 'That's just coincidence' — offered as a bare label, with no assessment of how improbable E would be if chance alone were operating, and no rival mechanism proposed. Conclusion 'no causal connection' is asserted without being earned.

Why it fails: It is the mirror image of jumping to a causal conclusion: instead of over-reading weak evidence, it under-reads strong evidence by invoking 'chance' as a stopping point rather than a hypothesis to be tested. Genuine coincidence is itself a probabilistic claim — it says the pattern is exactly what you'd expect from randomness — and that claim needs support (a base rate, a comparison class, an account of why the timing/mechanism argument is weak). Appeal to coincidence skips that work; it uses the word 'coincidence' as a conversation-ending shrug rather than a competing explanation, so the dismissal is unearned regardless of whether the underlying causal claim happens to be true.

• How probable would this exact pattern (timing, repetition, location, magnitude) be if pure chance were really the only thing going on?
• Has anyone laid out a comparison — chance hypothesis vs. proposed mechanism — or has 'coincidence' just been declared?
• Does calling it a coincidence conveniently make an uncomfortable finding, cost, or responsibility disappear?
• If the roles were reversed (the pattern favored the speaker), would 'coincidence' still be the first word reached for?

Legitimate when: When the correlation actually is weak, the sample is small, base rates of the 'coincidental' event are high, or a specific accounting shows the pattern is unremarkable under chance — refusing to infer causation there is sound skepticism, not this fallacy. The line is whether the chance explanation is argued for or merely invoked to end the discussion.

Built from: Wikipedia — Slothful induction · Logically Fallacious — Appeal to Coincidence

Appeal to Common Belief

also known as: argumentum ad populum, appeal to belief

A lot of people believing something doesn't make it true.

At the gym, your friend hands you a tub of powder: 'Everyone in this spin class takes creatine before a ride, so it must actually help with spin class.' You look around and, sure enough, half the room has a shaker bottle. But nobody in that room ran a test comparing riders who took it to riders who didn't. All you've actually learned is that the powder is popular in this room, not that it does anything for your ride.

The tell: The nagging sense that someone just counted heads instead of checking facts — you're being handed a headcount where you expected a reason.

The move: Let P be the claim in question and G a population — everyone, or the members of some named group. The arguer establishes (or merely asserts) that most of G believes P or acts as though P, and substitutes that frequency for the grounds P would need: most of G believes/does P; therefore P is true, correct, or the thing to do. The headcount over G is placed where evidence bearing on P itself belongs, and no premise is supplied connecting G's rate of belief to whatever would make P true. Because the substitution happens at that seam, the argument survives intact even when the popularity figure is perfectly accurate and P is false.

Why it fails: The number of people who hold a belief or follow a practice is a fact about the belief's popularity, not a fact that bears on the belief's truth. A claim about the world (what caused the flood, whether a drug works, what year an event happened) is settled by evidence connected to that claim, not by a headcount of opinions, and beliefs can become widespread for reasons entirely unrelated to their truth: social conformity, marketing, fear, repetition, or tradition. The move substitutes 'many people accept this' for the actual grounds the claim would need, so the conclusion can be false even when the premise about popularity is completely accurate.

• Is this the kind of claim that popular belief could actually constitute or reliably track (a word's meaning, a social custom) or is it a fact about the world that agreement cannot create (a physical cause, a historical date, whether a treatment works)?
• How was 'most people believe/do this' actually established — a real survey of a defined group, or just an unverified impression ('everybody knows')?
• If you subtract the headcount entirely, is there any independent evidence left standing for the claim itself?
• Would this same style of reasoning have justified beliefs that were once widely held but turned out false?

Legitimate when: Widespread belief or practice is legitimate evidence, not proof, in two narrow settings: (1) when the matter is itself constituted by collective agreement, such as correct grammar, word meanings, social conventions, or etiquette, where what most speakers/users do simply is the standard; and (2) as a weak, defeasible signal that something merits a closer look, when the 'crowd' is actually the relevant expert or evidentiary community (near-unanimous scientific consensus, for instance) rather than the general public — but that case is properly an appeal to authority riding alongside the popularity, and even then it only shifts a prior, it never substitutes for the underlying evidence.

Ancient: Cicero treats the agreement of all nations — that the gods exist and that something of us survives death — as weighty evidence, the consensus gentium argument that later critics would cite as the canonical appeal to common belief. — Cicero, Tusculan Disputations, Book I

Built from: Wikipedia · Internet Encyclopedia of Philosophy · Stanford Encyclopedia of Philosophy

Appeal to Common Folk

Someone says 'I'm just like you' and expects that to count as a reason to agree with them, instead of actually making their case.

A new financial advisor posts a video: 'I grew up broke. I shopped at the same stores you do, I clip coupons, I get it. So when I tell you to put your savings into this fund, you can trust me — I'm not some Wall Street guy in a suit.' He never explains the fund's fees, its track record, or its risks. When a viewer asks about the expense ratio, he replies, 'Look, I'm one of you, would I steer you wrong?'

The tell: The conversation is about whether the advice is good, but the answer you keep getting is about who the person is and how relatable they are, not what the advice actually does.

The move: S performs or claims membership in the audience's own group (ordinary, plain, working, one-of-us); S asserts P; therefore P should be accepted, or a rival view should be rejected because its holder is not one of us. Converse form: C objects to P; C is marked as an outsider/elite/not-plain-folk; therefore C's objection can be dismissed.

Why it fails: Shared identity, background, or manner is offered as if it were evidence for the truth of P, but being ordinary confers no special access to facts about tax policy, trade deals, medicine, or whatever the claim concerns; at most it might show the speaker shares the audience's interests, which is a different question from whether the claim is true. The move also runs backward as a shield: framing an opponent as an out-of-touch elite lets their argument go unanswered because the person, not the claim, has been disqualified.

• Is the speaker's ordinariness (habits, accent, job, background) being offered as a reason to believe the claim itself?
• Does the shared background actually confer relevant knowledge or evidence on the specific question at issue, or just shared taste and grievance?
• If an outsider or an 'elite' made the identical claim with the identical evidence, would it be received any differently?
• Is a critic being dismissed as out-of-touch instead of having their actual objection addressed?

Legitimate when: Common experience is genuine evidence when the claim is specifically about what ordinary life is like for people in that position (a factory worker's testimony about factory conditions, a renter's testimony about what renting is actually like), and the identity claim is offered alongside a substantive case rather than as a replacement for one.

Ancient: Marius, a 'new man' with no noble ancestry, tells the crowd that the aristocrats have only read or heard of the things he has personally lived and fought through, that his scars are his statues and his nobility, and that he will march and camp alongside the common soldier just as one of them rather than living at ease while they suffer — urging the people to trust his plain, unpolished manner over the nobles' refined words, since (he implies) fine speech is what men use to dress up bad character. — Gaius Marius's speech to the Roman assembly, as reconstructed by Sallust, Bellum Jugurthinum (The Jugurthine War), ch. 85

Built from: Wikipedia — Plain folks · Logically Fallacious — Appeal to Common Folk · Propaganda Critic — Plain Folks · Stanford Encyclopedia of Philosophy — Fallacies (ad populum) · Wikipedia — List of fallacies (argumentum ad populum)

Appeal to Common Sense

Calling something 'just common sense' so nobody has to explain it or check if it's actually true.

At a kid's birthday party, one parent is quietly keeping their son away from the cake table. A grandparent scoffs: 'Oh, come on, everyone knows sugar makes kids bounce off the walls, it's just common sense.' Another parent mentions that when people have actually run careful tests -- where nobody knew which kids got sugar and which didn't -- the bouncing didn't show up any more than usual. The grandparent waves it off: 'You can't argue with common sense. Look at any kid after cake.'

The tell: You ask a plain 'how do we actually know that?' and instead of an answer you get impatience, the same line repeated louder, or a jab that you must be overthinking something 'everybody' already knows.

The move: P is asserted to be 'common sense' or obvious to any sensible person; therefore P is true, and further inquiry into whether P is actually true can be waived.

Why it fails: Calling a claim 'common sense' reports how comfortable or familiar it feels, not why it is true. That report is not evidence for the claim itself, and the move is deployed precisely on the claims where intuition and reality are known to part ways -- causal mechanisms, base rates, incidence, anything counter-to-experience at small or large scale. The defect is not that shared background knowledge is illegitimate as a consideration; it is that 'it's obvious' is offered as a terminal stopping point for a claim that is actually checkable, and the appeal is used specifically to foreclose the check rather than to summarize one already done.

• Is 'common sense' here pointing to a shared, checkable observation, or just asserting that the claim feels obvious?
• Is the claim actually testable, and has anyone tested it -- or is the appeal being used specifically to avoid running that test?
• Would this seem equally obvious to someone without the speaker's particular background, incentives, or first impression?
• Has 'common sense' of this same kind been wrong before in a comparable case?

Legitimate when: Appeal to shared background knowledge is legitimate when the claim really is a settled, low-stakes observation available to any competent observer (e.g. 'don't grab the pan handle, it's hot'), no one has offered a specific, checkable reason to doubt it, and the appeal is pointing to evidence everyone already has rather than substituting for evidence that could be gathered.

Ancient: Confronted with Zeno's arguments that motion is impossible, Diogenes said nothing in reply -- he simply stood up and walked, treating the plain fact of moving as too obvious to need an answer to the argument itself. The anecdote is often admired as a gesture, but it does not identify what is wrong with Zeno's reasoning; it substitutes the obviousness of the conclusion for engagement with the argument that seemed to threaten it. — Diogenes the Cynic, per the later report of Simplicius of Cilicia

Built from: Logically Fallacious · Wikipedia -- Argument from incredulity · Wikipedia -- Solvitur ambulando

Appeal to Complexity

Someone waves off a specific finding by declaring the whole topic 'too complicated for anyone to really understand' — instead of checking whether the finding is actually right.

In the family group chat, Priya posts a screenshot showing Uncle Ray's phone bill has been overcharging him the same fee for two years — she circled the exact line item. Ray replies, 'Phone plans are a total maze, honestly nobody outside the company understands how they price this stuff, so I wouldn't put too much stock in your math.' He never opens the screenshot or points to what's wrong with it.

The tell: The person never touches the actual number or claim in front of them — they just declare the whole subject too big and messy to deal with, and somehow that same 'too messy to know anything' problem never seems to stop them from confidently sticking with their own original answer.

The move: Subject X is very complex (many interacting parts, technical, hard to follow in full). Therefore no one — including the person presenting a specific, narrow claim about X — can really know claims about X. Therefore the specific claim/finding about X should be rejected (or the arguer's own simpler alternative should be preferred) without examining its actual content.

Why it fails: Complexity is a fact about how much work is needed to master an entire domain; it says nothing about whether one particular, bounded claim within that domain has been checked and is true or false. The arguer treats their own (or an imagined audience's) failure to grasp the whole system as proof that no correct specific claim can be made within it, ignoring that expertise, narrowed scope, and independent verification exist precisely to handle complexity. It is a non sequitur that substitutes 'hard for me to verify quickly' for 'impossible for anyone to know,' and the standard is applied asymmetrically: the arguer's own preferred conclusion, resting on the same complex subject, is quietly exempted from the very impossibility just invoked against the opponent.

• Does the subject's overall complexity actually block the specific, bounded claim being disputed, or does it just make the whole topic harder to survey?
• Is the arguer's personal difficulty following the analysis being generalized into 'nobody can know this'?
• Has anyone with relevant expertise actually examined the specific claim, or is it being dismissed before any specialist review happens?
• Would the same complexity standard, honestly applied, also disqualify the arguer's own preferred conclusion or proposed alternative?

Legitimate when: Flagging complexity is legitimate when it is used as a call for real verification or expertise before a claim is accepted ('this is intricate, let's get someone qualified to check it') rather than as the final word that ends the discussion; it is also legitimate when the complexity genuinely does exceed anyone's actual investigative access (destroyed records, classified data, systems no living person can inspect) — but then the same limit must bind every competing claim on the table, including the arguer's own, not just the one they want to reject.

Built from: Logically Fallacious · Wikipedia - List of fallacies · Wikipedia - Argument from ignorance · Internet Encyclopedia of Philosophy - Fallacies

Appeal to Consequences

also known as: argumentum ad consequentiam

Deciding something can't be true just because it would be a bummer if it were.

In the group chat, someone posts an article claiming the bakery on 5th Street uses artificial vanilla flavoring instead of real vanilla. One friend fires back: 'No way that's true. Friday cupcake runs are the best part of my week -- if that were real, I'd have to stop going.' Nobody has checked the ingredient list. The friend has only established that the story would be annoying, not that it's wrong.

The tell: You notice the other person is telling you how bad it would be if the thing were true, instead of telling you any reason to think it isn't.

The move: Negative form: If P were true, consequence C would follow; C is unwelcome (undesirable, frightening, embarrassing); therefore P is false. Positive form (mirror image): If P were true, consequence C would follow; C is welcome; therefore P is true. In both directions the argument swaps a judgment about how good or bad C would be for a judgment about whether P actually obtains, without touching any evidence that bears on P itself.

Why it fails: Whether P is true is settled by how the world is, not by how we feel about what follows if it is. The desirability or awfulness of a belief's implications is not evidence for or against that belief's truth; the arguer has answered 'would I want this to be so?' while pretending to have answered 'is this so?'. The negative form also mimics modus tollens (if P then C; not-C; therefore not-P) closely enough to feel valid, but 'C is undesirable' is smuggled in where 'C is false' is required -- disliking C is not the same as C failing to obtain.

• Is the unwelcome (or welcome) consequence being used against the claim's truth, or against a proposed action -- where weighing consequences is exactly the right tool?
• Does the goodness or badness of the consequence change any of the actual evidence for or against the claim, or does it only change how we'd feel if the claim held?
• Would the arguer accept the mirror-image move -- 'this would be pleasant to believe, therefore it's true' -- or do they only run the inference in the direction that favors their preferred conclusion?
• Strip out the emotional weight of the consequence: does the argument still have anything left connecting the premise to the claim's truth?

Legitimate when: Reasoning from consequences is the correct tool whenever the question on the table is what to do rather than what is true: choosing between policies, plans, laws, or actions is properly settled by weighing their likely outcomes (the core move of consequentialist ethics, cost-benefit analysis, and the legal doctrine of argumentum ab inconvenienti, where courts legitimately reject an interpretation of a rule because it would produce absurd or unworkable results in future cases). It is also legitimate when the 'consequence' is a further, already-accepted proposition that P strictly implies -- there, an unacceptable logical consequence really is evidence that P is false (genuine reductio ad absurdum), because the connection runs through logical entailment rather than through what we'd prefer.

Built from: Wikipedia · The Fallacy Files (Appeal to Consequences) · Logically Fallacious · Internet Encyclopedia of Philosophy (Fallacies)

Appeal to Definition

Someone thinks looking a word up in the dictionary settles an argument that was never really about the word.

At book club, Dana insists the novel everyone loved isn't really a 'thriller' because her dictionary app says a thriller needs 'a crime and physical danger,' and this book's tension is all emotional. The group stops talking about whether the book actually gripped them and spends ten minutes arguing over the dictionary entry instead.

The tell: The conversation quietly shifts from the real question you cared about (was the book gripping? is this fair? should this be allowed?) to a side argument about which definition is the correct one.

The move: Definition D of term T (drawn from a dictionary, a policy document, or one accepted usage) is cited; the case or claim c satisfies (or fails) D's wording; therefore the substantive question about c — a matter of fact, law, policy, or value — is thereby settled, without independently establishing that D is the definition actually relevant to deciding c.

Why it fails: General-purpose definitions are usually descriptive summaries of common usage, not stipulative rulings that can arbitrate a contested factual, legal, or policy question. A single word routinely carries several legitimate senses (colloquial, technical, legal, contextual), and which sense governs the disputed case is exactly what is at issue — so citing 'the' definition quietly presupposes the answer instead of arguing for it. The move also invites cherry-picking: of several dictionary or source definitions, the one selected tends to be whichever happens to support the arguer's conclusion, while a domain-appropriate or purpose-sensitive sense that would decide the case differently is ignored.

• Is the disputed question actually a matter of fact, law, policy, or value — or is it merely a question of how a word is ordinarily used?
• Which definition is being cited, from what source, and was it selected because it happens to support the conclusion (i.e., could an equally ordinary definition have been chosen instead)?
• Does the source of the definition (a general dictionary, a lay glossary) have any authority over the specialized, legal, or contextual domain the argument is actually about?
• If a different, equally legitimate definition of the same term were substituted, would the argument's conclusion survive?

Legitimate when: Citing a definition is legitimate when the dispute really is about linguistic usage or meaning (clearing up a genuine misunderstanding of what a word denotes) or when the definition invoked is the one the governing framework itself stipulates as authoritative for the matter at hand — a contract's or statute's own definitions section, an accepted technical standard within the relevant field, or a term the parties have already agreed to use in a specific sense. It fails when a definition is used to short-circuit a substantive disagreement that the wording alone cannot decide.

Built from: Logically Fallacious · Effectiviology · Wikipedia — No true Scotsman

Appeal to Desperation

Someone pressures you into backing their plan just because doing SOMETHING feels better than doing nothing — without ever showing their plan actually works.

Your apartment's pipes have been dripping for weeks and now there's a water stain spreading on the ceiling. Your roommate says, 'We have to do something right now — I already ordered a $400 emergency pipe-wrap kit from an ad I saw, it's arriving tomorrow.' You ask whether the kit even fixes a ceiling leak. They snap back, 'Oh, so you have a better idea? No? Then we're doing this.' The kit turns out to be insulation for outdoor faucets in winter weather — it does nothing for an indoor ceiling leak. The actual problem never gets looked at, because the panic about the leak was used to shut down every follow-up question about the fix.

The tell: The pressure is all about how scary or urgent the problem is, never about any evidence that this particular fix helps — and the person gets defensive or annoyed the moment you just ask 'how would that actually solve it?'

The move: Something must be done about problem P. X is something (a candidate action). Therefore, X must be done — and questioning X is treated as equivalent to endorsing inaction on P.

Why it fails: The move trades on an equivocation between 'something' as an unspecified existential claim (some action or other is warranted) and 'this specific something' (X). That a crisis demands some response is real, but it supplies no evidence that X in particular is effective, targeted, or superior to the available alternatives — the urgency premise and the merit of X are simply not connected. It is a relevance failure: desperation is offered as though it were evidence for X's soundness, when at most it supports the disjunctive claim that some action is called for. The move also typically closes off scrutiny by recasting any question about X as complicity with the crisis, which is a separate strawman layered on top of the core non sequitur.

• Does the urgency of the problem actually show that THIS proposal, specifically, will work — or only that some action is needed?
• Has this proposal been compared against other candidate actions, or was it accepted only because it was the one already on the table?
• Is 'we have to do something' being used to make ordinary scrutiny of the proposal look like rooting for the crisis to continue?
• Strip away the time pressure — is there any actual evidence this specific action addresses the cause of the problem, or just that it addresses the anxiety about the problem?

Legitimate when: When a deadline is genuinely real and waiting for certainty is itself the costlier option, choosing a reasonable (even imperfect) action under uncertainty can be justified — but only when the proposal is selected because it is the best available option given the constraint, not merely because it is the only one voiced, and only when at least some minimal reason for that specific choice is offered rather than urgency alone standing in for the reason.

Built from: Wikipedia — Politician's syllogism · Logically Fallacious — Appeal to Desperation

Appeal to Emotion

Someone makes you feel something strong on purpose, then hopes that feeling will do the arguing for them instead of any actual reason.

A friend group chat is deciding where to eat. One person doesn't want the vote to go to the new salad place, so instead of saying why, they type: 'Wow, guess nobody cares that I've had the worst week of my life and just wanted one nice comfort meal with people who love me.' Nobody answers the actual question of where to eat. Everyone just scrambles to comfort them, and the group ends up at the comfort-food spot.

The tell: You notice the conversation has stopped being about the actual question and has turned into managing someone's mood, guilt, or fear. You feel a tug to agree just to make the bad feeling stop, even though you couldn't explain, in plain words, what reason you were just given.

The move: An emotion E (fear, pity, anger, joy, outrage, sympathy, etc.) is aroused in the audience by some description, image, or appeal; the intensity of the felt E is then substituted for evidence that claim P is true or that action A should be taken; therefore P / therefore A. The emotional charge, not any fact connecting E to P, is what does the persuasive work.

Why it fails: The strength of a feeling is not itself evidence for the truth of a proposition. When the arousing detail has no probative connection to the matter actually at issue, the argument commits a relevance failure (a red herring): it offers a psychological cause of belief in place of a reason to believe. An audience can feel the aroused emotion just as strongly whether P is true or false, so the emotion tracks nothing about P's truth-value; it only tracks how the material was presented.

• Which specific emotion is being aroused, and is it tied to any actual fact or reason bearing on the claim, or does it stand free of one?
• Would the argument retain any force for a listener who feels nothing but is still paying attention to the facts?
• Are the emotion-evoking details (the suffering child, the vivid threat, the flattering comparison) themselves relevant evidence about the matter being decided, or are they imported purely for their emotional charge?
• If you strip out every emotionally loaded word and image and restate the claim in flat, neutral language, does any support for the conclusion survive?

Legitimate when: Emotion-evoking facts are not automatically illegitimate: an emotion is properly probative when the very thing that makes it distressing, pitiable, or alarming is also a genuine reason bearing on the decision at hand (e.g., a victim-impact statement describing real harm is relevant to a sentencing decision; a vivid but accurate description of a risk is relevant to a safety decision). The move becomes fallacious specifically when the emotional charge is doing work that no accompanying fact could do on its own, or when it is used to override, distract from, or substitute for evidence that would otherwise settle the question.

Ancient: Aristotle criticizes earlier handbook writers for teaching speakers how to arouse a judge's prejudice, pity, or anger rather than sticking to the facts of the case, comparing the tactic to warping the measuring-stick before you use it to measure — even as his own Rhetoric goes on to give a systematic account of how the emotions work, on the view that a speaker should understand them well enough to know when they are legitimately in play. — Aristotle, Rhetoric

Built from: Wikipedia · Stanford Encyclopedia of Philosophy · Internet Encyclopedia of Philosophy

Appeal to Equality

Saying two things are equal in one way, then acting like that settles a completely different question about how they should be treated.

Two roommates have always split the grocery bill straight down the middle because they're 'equal roommates.' One month, one of them eats almost every meal out and barely opens the fridge, while the other cooks dinner for both of them nearly every night. When the second roommate suggests splitting that month's bill differently, the first one cuts them off: 'We're equal roommates, so we split everything equally. That's just fair.'

The tell: Someone leans hard on the word 'equal' or 'fair' but nobody has actually checked whether the two situations match on the specific thing that's being decided — only on some other, more convenient thing.

The move: A and B are equal in respect R1; fairness requires treating equals equally; therefore A and B must be treated identically in respect R2, though R2-relevant differences are unaddressed.

Why it fails: The equality that actually holds (R1) is quietly substituted for the equality that would be needed to justify the conclusion (R2). The norm 'treat equals equally' is only valid when the equals in question are equal along the dimension the treatment tracks; here that has not been shown, only presumed. The move is a double failure: it borrows the leveling structure of false equivalence (one shared trait is inflated into total sameness) and the structure of the fallacy of accident (a generalization or standard is applied without regard to an exception that is precisely what's at issue).

• In which specific respect are the parties or cases actually equal, and is that the same respect the proposed treatment depends on?
• Are there relevant differences between the parties that the equal-treatment demand is ignoring or waving past?
• Does the governing fairness standard here actually call for identical treatment, or for proportionate treatment given the differences that exist?
• If the 'equal' framing were dropped and the parties were compared feature-by-feature on the point that matters, would the equality survive?

Legitimate when: When the respect in which the two parties are shown to be equal is the very respect that the proposed treatment turns on, and no material difference bears on that specific dimension — so that proportionate treatment and identical treatment coincide.

Ancient: Aristotle observed that a feature true of a thing only incidentally — true of it 'in a certain respect' rather than in itself — gets wrongly carried over as though it held without qualification. Something equal or true in one limited respect is treated as if it were equal or true across the board, which is the same slide this trick relies on: sameness that is only accidental gets promoted into sameness that would actually justify the conclusion. — Aristotle, discussed in the Stanford Encyclopedia of Philosophy's account of the fallacy of accident (Sophistical Refutations)

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — List of fallacies · Wikipedia — False equivalence · Internet Encyclopedia of Philosophy — Fallacies · Logically Fallacious — False Equivalence · Wikipedia — Fallacy of relative privation

Appeal to Extremes

Someone blows up your small, careful idea into a wild extreme, knocks down the extreme, and acts like they beat your actual idea.

You tell your roommate, 'Maybe we should rinse out the cardboard boxes before recycling them.' They shoot back, 'Oh sure, and next you'll want us digging through the neighbor's trash bins and building a compost pile in the living room. Let's just forget the whole recycling thing.' You never said any of that — but now the conversation is about compost piles instead of cardboard boxes.

The tell: That flash of 'wait, that's not even close to what I said' — the version they're arguing against feels like a cartoon of your actual point, way bigger and weirder than what you asked for.

The move: Speaker A asserts a moderate, bounded claim P. Speaker B restates P as P* — an unqualified, extreme version of P that A never asserted and that does not follow from P without an unstated bridging premise. B then shows P* is absurd or unacceptable and treats this as having refuted P itself: 'If P, then (absurd extreme) P*; P* is unacceptable; therefore P is false / should be rejected.'

Why it fails: The extrapolation from P to P* is asserted, not demonstrated: no entailment, no shared limiting principle, and often an explicit qualifier in P is silently dropped to manufacture P*. Because P* is a different, stronger claim than P, refuting P* leaves P's actual truth or merit untouched — the argument commits a relevance failure (it answers a claim nobody made) dressed up as if it disposed of the real one.

• Does the extreme version actually follow from the claim as its proponent qualified it, or was a limiting word ('some,' 'in school zones,' 'during meetings') quietly dropped?
• Has the arguer engaged the original, moderate claim anywhere, or only ever the stretched-out version?
• Is there a stated or obvious boundary in the original proposal that would stop it from sliding to the extreme — and did the response account for that boundary?
• If you grant the extreme conclusion is absurd, does that absurdity actually transfer back onto the modest claim, or only onto the invented stand-in?

Legitimate when: The extrapolation is legitimate reductio ad absurdum when the extreme case truly is entailed by the claim as stated — i.e., no limiting principle in the claim blocks it, and the arguer can show the logical bridge from P to P* rather than merely asserting it. In that case refuting P* really does refute P.

Ancient: Aristotle's catalogue of ways an argument can go wrong outside the words themselves includes answering a claim different from the one actually raised. Later logic treats the trick of swapping in an exaggerated stand-in for an opponent's real position as a variant of that same family. This is the lineage the move descends from, not a surviving classical example of the extreme-extrapolation trick itself — no ancient source spells out this exact stretched-to-the-extreme pattern by name, so this attribution is offered as ancestry, not a direct hit. — Aristotle, Sophistical Refutations (De Sophisticis Elenchis) — the broader category of 'missing the point' / irrelevant-conclusion reasoning that later writers used to name this specific tactic

Built from: Logically Fallacious — Appeal to Extremes · Wikipedia — Straw man · Wikipedia — Slippery slope · Wikipedia — List of fallacies · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies

Appeal to Faith

Saying "I just believe it" instead of answering the question, and then acting like not believing it too is the real problem.

A friend group chat is arguing about whether a landlord will actually return the security deposit. One person keeps saying, "He's not going to give it back, I can already tell." Someone asks, "Did you check the lease, or text him?" She replies, "I don't need to check anything, I just have a feeling about this guy, and honestly I'm surprised you're being so trusting." Nobody has looked at the lease. Nobody has texted the landlord. The conversation has quietly become about whether you're naive for wanting to check.

The tell: You asked a question that has an actual answer -- something you could look up or check -- and instead of an answer you got a feeling, plus a jab about your character for asking.

The move: A claim P is put forward and questioned. Instead of supplying evidence or argument for P, the arguer states that P is accepted on faith (in P itself, or in the person/institution asserting P), and treats 'I have faith' as if it settled the matter. When the questioner persists, the demand for evidence is recast as a lack of faith or loyalty, shifting the burden from the arguer (to support P) onto the questioner (to justify doubting it).

Why it fails: Stating that one believes something regardless of evidence does not supply any reason for a third party to believe it too; faith is a report of the arguer's psychological stance, not a consideration that bears on whether P is true. Where P is a claim of the kind ordinarily settled by observation, testimony, or inspection (e.g., 'the water tested safe'), invoking faith in place of the available checks does not merely fail to support P -- it substitutes an unfalsifiable assurance for the specific evidence the audience needs, and the closing move of branding the doubt itself as disloyalty adds a second failure (deflecting scrutiny rather than answering it) on top of the first.

• Is faith being offered as the reason to believe P to someone who does not already share that faith?
• Is P the kind of claim that has an available, checkable answer (a test, a record, a demonstrable fact) that the appeal to faith is being used to bypass?
• Does the response reframe the questioner's request for evidence as a personal failing -- disloyalty, cynicism, lack of belief -- rather than addressing the request?
• Would 'I have faith in P' count as a reason for literally any claim P, no matter how implausible -- and if so, what is it actually doing in the argument?

Legitimate when: It is not this fallacy when faith is offered as a report of one's own commitment rather than as evidence meant to persuade someone else of a checkable fact -- e.g., 'I have faith this treatment will help me get through chemo' explains a person's private resolve, not a claim about the treatment's clinical efficacy. It is likewise not this fallacy when the matter under discussion is genuinely outside the reach of the available evidence (a properly acknowledged article of faith, openly presented as such, in a context -- such as theological discussion among people who share the framework -- where it is not being used to settle an empirical dispute or silence a legitimate question).

Ancient: Tertullian writes that the death of the Son of God is credible precisely because it is unfitting ("credibile est, quia ineptum est"), and that the burial and resurrection are certain because impossible ("certum est, quia impossibile") -- the passage later popularized, and distorted, as "credo quia absurdum" ("I believe because it is absurd"). — Tertullian, De Carne Christi 5

Built from: Stanford Encyclopedia of Philosophy - Fallacies · Logically Fallacious (Bo Bennett) - Appeal to Faith · Wikipedia - Credo quia absurdum

Appeal to False Authority

also known as: argumentum ad verecundiam, false authority

Someone sounds official or has an impressive title, so people believe what they say — even though that title has nothing to do with what they're actually claiming.

In a family group chat, your uncle shares a post: "My dentist says the new sugar substitute in diet soda causes cancer — she's literally a doctor, trust her." Within an hour three relatives have thrown out their soda. Nobody asks what a dentist's training in teeth has to do with judging cancer research on food chemistry, or whether actual toxicologists have looked at the substitute at all.

The tell: You notice yourself nodding along because of the person's job title or credentials, not because you actually followed or checked their reasoning.

The move: Person A asserts that P is true. A is presented (or treated by the audience) as an authority whose say-so counts as evidence for P. But A is not actually a competent authority on P — A lacks expertise in the specific domain of P, has a conflicting interest, or has been misquoted or quoted out of the domain in which A actually has standing. Therefore, P is true.

Why it fails: The move trades on the appearance of expertise rather than expertise that actually bears on the claim at hand. Testimony from a genuine, relevant, disinterested authority can be real evidence, but here the authority is either not an authority at all, not an authority on this particular question, is contradicted by the relevant expert community, or is not accurately reporting what they said — so the stated reason gives no real support to the target claim even though it is presented as if it settles the matter.

• Is the person's claimed expertise actually in the specific domain of this claim, or in some adjacent or unrelated field?
• Would recognized specialists in the relevant field broadly agree with this person's claim, or is this person a outlier relative to expert consensus?
• Is the person being quoted accurately and in the context they said it, or has the statement been simplified, cherry-picked, or lifted from a different domain?
• Does the person have a financial, professional, or personal stake in the claim being believed, independent of their expertise?

Legitimate when: Citing an authority is legitimate when the person genuinely holds current, relevant expertise in the specific domain of the claim, when that view is representative of (not an outlier within) expert consensus, when the statement is reported accurately and in context, and when the audience treats the testimony as defeasible evidence open to revision rather than as proof that ends inquiry.

Ancient: Locke describes the move of citing men esteemed for learning, eminence, or power and treating any disagreement with them as an act of insolence, so that the listener's deference to the person's standing — rather than any evidence — is what carries the point. He names this tactic argumentum ad verecundiam, appeal to modesty. — John Locke, An Essay Concerning Human Understanding, Book IV (the chapter on reason)

Built from: Wikipedia — Argument from authority · Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — List of fallacies

Appeal to Fear

Someone tries to scare you into agreeing instead of actually giving you a reason.

You're buying a laptop and you tell the salesperson you'll skip the extended warranty. He leans in and says, 'Your call, but hard drives fail all the time. Picture it: every photo of your kids growing up, gone, no warning, right when you need it least.' He never mentions how often this laptop's drives actually fail, what the warranty even covers, or what it costs to just back up your files yourself for five dollars a month. He just needs you to picture the empty photo album.

The tell: The conversation stops being about facts and starts being about a scary picture in your head. You feel a jolt of dread, but if someone asked you to explain out loud exactly why saying no would actually cause the bad thing, you'd realize you can't.

The move: A frightening outcome F is presented as what will happen if claim P is rejected or action A is not taken; the vividness and imminence of F, rather than any demonstrated causal or evidential link between rejecting P and F occurring, is what carries the audience to the conclusion. Schematically: 'If not-P, then (frighteningly) F; F must be avoided; therefore P.' Often paired with a false dilemma that hides every option except the frightening one and the arguer's preferred one.

Why it fails: Being afraid of F is not evidence that rejecting P causes F, or that P is true. The argument needs two things it never supplies: that F is actually likely to follow from not-P, and that the magnitude of F (as opposed to just its emotional charge) has been honestly weighed against the cost of accepting P. When the frightening image substitutes for that missing causal and probabilistic case, the audience is moved by dread rather than by any reason connecting the two claims — the fear is doing persuasive work that only evidence should do.

• Is the feared outcome actually caused by rejecting the claim, or is it just placed next to the claim so the two feel connected?
• Would the argument still persuade if you removed the vivid imagery and stated the plain probability of the harm instead?
• Is the likelihood of the feared outcome ever quantified, or only evoked with words like 'imagine if' or 'picture this'?
• Is the speaker forecasting a harm from some outside cause, or actually threatening to inflict the harm themselves if you don't comply? (The latter is appeal to force, not appeal to fear.)

Legitimate when: Warning someone of a real, evidenced risk is not fallacious just because it is frightening — a doctor describing a surgery's documented complication rate, a structural engineer flagging a cracked support beam, or a weather service issuing an evacuation order are all appealing to genuine, quantified danger as one relevant factor among others. The move turns fallacious specifically when the fear is the only thing doing the persuading, the causal link to the rejected claim is asserted rather than shown, and the probability of the outcome is left vague on purpose because a clear number would weaken the pitch.

Ancient: Aristotle analyzes fear (phobos) as pain or disturbance arising from the impression that some destructive or painful evil is close at hand, and explains that a speaker can make an audience feel this by making the danger seem near and the audience seem vulnerable to it. His treatment is descriptive rather than a condemnation — he is cataloguing how the emotion works so a speaker can understand and use it — but it is the ancient source later fallacy theorists point to when they identify fear as a lever an arguer can pull independent of the actual merits of a case. — Aristotle, Rhetoric, Book II

Built from: Wikipedia · Stanford Encyclopedia of Philosophy · Logically Fallacious

Appeal to Flattery

Someone showers you with compliments right before asking for something, so that saying no feels like you're proving the compliment wrong.

In the group chat about the family reunion, Jake writes: 'You all have the best taste of anyone I know, seriously, people always say that about this group — so I already booked the lake house I found, no need to even look at other options, right?' Nobody wants to be the one who 'doesn't have good taste,' so the venue goes unchallenged.

The tell: You notice the praise landed right before the ask, not because anyone actually cares about your taste, your smarts, or your strength — it showed up exactly when they needed your yes.

The move: You (the audience) possess admirable trait Q; people who possess Q would naturally accept/do P; therefore accept/do P. The compliment about Q is offered as if it were a reason to accept P, when it has no bearing on whether P is true or wise.

Why it fails: Praise of the listener is not evidence about the proposition being urged. The flattery works by attaching a social cost to disagreement — refusing now feels like disproving the compliment ('surely someone as smart as you can see this') — rather than by giving any reason P is actually correct or advisable. Because the same compliment could be pinned onto literally any conclusion with equal force, it carries zero evidential weight; its only work is psychological pressure, and per Wikipedia it is 'usually a cunning form of appeal to consequences, since the audience is subject to be flattered as long as they comply.'

• Would this exact compliment work just as well glued onto the opposite conclusion?
• Does declining the request actually contradict the flattering quality being praised, or is that link only implied to create pressure?
• Strip the compliment out entirely — is there any argument left standing on its own?
• Did the praise arrive independently of the ask, or does it appear right at the moment agreement is being sought?

Legitimate when: The praise is sincere, independent of the request, and not doing the work of persuasion — or the complimentary statement is itself a genuinely relevant fact about the matter at hand (e.g., telling someone their strong measurements make them a good fit for a modeling job is relevant to that specific proposal, not a substitute for reasons).

Ancient: Socrates classes rhetoric together with cookery as forms of flattery (kolakeia): knacks aimed at gratifying an audience rather than at truth or the good, producing pleasure in the listener in place of genuine grounds for belief. — Plato, Gorgias

Built from: Wikipedia — Appeal to flattery · Wikipedia — List of fallacies · Logically Fallacious — Apple Polishing · Internet Encyclopedia of Philosophy — Fallacies

Appeal to Force

also known as: argumentum ad baculum, appeal to the stick

Someone gets you to agree not because they're right, but because they can hurt you if you don't.

You tell your roommate the cleaning schedule they wrote up isn't fair because it has you doing dishes every single night. Instead of explaining why the schedule makes sense, they say, 'Fine, keep arguing about it, and see if I don't tell the landlord it was your idea to sneak in a second cat.' Nothing about that response tells you the schedule is actually fair. It just tells you what happens if you keep pushing.

The tell: You notice the conversation has quietly stopped being about who's right and started being about what you stand to lose. The other person hasn't answered your point at all, they've just made it more expensive to keep making it.

The move: The arguer holds (or claims to hold) the power to inflict some harm T on the audience, and presents acceptance of claim P as the only way to avoid T: 'Accept P, or I will do T to you; therefore P.' The threatened harm is offered in the place a reason for P's truth should occupy, and the harm's source is the speaker's own will, not an independent fact about the world.

Why it fails: A threat can give someone a genuine reason to act a certain way, but it cannot give anyone a reason to believe a certain way. Whether P is true or false does not change depending on who is willing to hurt you over it, so a threat directed at belief supplies zero evidential weight for P — it only changes the cost of disagreeing. The fallacy specifically targets doxastic appeals ('believe P or else') rather than practical ones ('do X or else'), and it is aggravated when the harm is one the speaker themselves will inflict, since that removes even the indirect informational value a third-party warning might carry.

• Is the threatened consequence caused by the truth of the matter, or is it simply the threatener's own choice to retaliate?
• Would the conclusion retain any support at all if the speaker suddenly lost the power to carry out the threat?
• Is the coercion aimed at what you should do (a practical choice, where consequences are legitimate reasons) or at what you should believe (a factual question, where they are not)?
• Remove the threat entirely and read only the remaining claim: is there any actual evidence left, or just the bare assertion?

Legitimate when: Consequences are legitimate reasons in practical deliberation: a union that says 'settle fairly or we strike,' a nation that says 'withdraw or we cut off trade,' or an employer who says 'wear the hard hat, it's state law and the job requires it' are all pointing to real stakes that bear directly on what someone should decide to do. It only becomes the fallacy when the same threat is redirected at belief itself — 'accept that X is true, or else' — or when the stated 'reason' is nothing but the threat with no independent justification attached.

Ancient: The Athenian envoys tell the Melians that questions of justice arise only between equals in power, and that the strong do what they can while the weak suffer what they must — pressing Melos to submit on the basis of Athenian military strength rather than on the merits of Melos's claim to neutrality. — Thucydides, History of the Peloponnesian War, Book V (the Melian Dialogue)

Built from: Stanford Encyclopedia of Philosophy · Wikipedia · Logically Fallacious

Appeal to Heaven

Someone says God (or fate, or 'the universe') personally told them their plan is right, so you're not supposed to argue with it anymore.

Your uncle wants to spend the family inheritance money on a boat. When your sister asks to see the numbers, he says, "I prayed about it all week, and I just know in my heart this is what Mom would have wanted — God's telling me to go for it." Nobody at the table brings up the price of the boat again. How do you argue with that?

The tell: The conversation about the actual facts — the price, the risk, whether the plan even makes sense — just stops. Somebody pulled out a card nobody else at the table is allowed to see, let alone check.

The move: A divine or providential source wills/has decreed P; therefore P is true, right, or obligatory — with the invocation of that will offered as sufficient by itself to settle the matter, in place of any independently checkable reason.

Why it fails: The claimed divine will is doing all the argumentative work, yet it is not something the audience can verify, examine, or dispute by any shared method — no one but the speaker has access to it. Because the claim is unfalsifiable, it can be asserted with equal rhetorical force for a conclusion and for its exact opposite, so it fails to actually discriminate between correct and incorrect positions. It substitutes an unexaminable assertion for a relevant, checkable reason, and it forecloses further scrutiny of the real merits (costs, evidence, rules) by placing the conclusion beyond argument.

• Is the claimed divine sanction verifiable by any means available to the audience?
• Could the same divine-will claim be asserted with equal force for the opposite conclusion?
• Are independent, examinable reasons offered alongside the appeal, or does the appeal stand alone as the entire case?
• Who is claiming access to God's will here, and what would count as evidence that they are wrong?

Legitimate when: When 'God wills it' is shorthand for a claim inside a shared, examinable framework the audience already accepts — e.g., citing a specific scriptural passage within a community that treats that text as its agreed authority, so the real support can still be checked against the shared source — or when offered only as an autobiographical report of the speaker's own motivation ('I feel called to do X') rather than as proof that X is objectively correct or binding on anyone else.

Ancient: When Urban called on Western Christians to march east, the assembled crowd is said to have answered with the chant "Deus vult" — "God wills it" — which Urban then sanctioned as the crusaders' official war cry. The campaign's rightness was thereby proclaimed settled by divine decree, spreading as a battle-shout at Amalfi and Antioch, rather than argued on its practical or moral merits case by case. — Pope Urban II at the Council of Clermont, 1095 (as recorded later by Robert the Monk's Historia Iherosolimitana, c. 1120, and echoed in the anonymous Gesta Francorum, c. 1100)

Built from: Logically Fallacious · Wikipedia — Pine Tree Flag (redirect target for "Appeal to Heaven") · Wikipedia — List of fallacies · Wikipedia — Deus vult

Appeal to Intuition

Treating a gut feeling as proof, instead of showing why it's true.

In the group chat, Dana posts a spreadsheet showing the new catering vendor is $400 a month cheaper with better reviews. Marcus replies, 'I don't need to see the numbers, something about this vendor just feels off to me, I've got a good sense for these things.' Nobody pushes back. The group quietly keeps the old vendor -- and nobody, including Marcus, ever names a single actual reason.

The tell: Notice that nobody, not even the person having the feeling, can point to one concrete fact behind it -- the feeling is doing all the work, and it's being used to shut the numbers out rather than add to them.

The move: Form A (rejection): There is evidence for X. X conflicts with the arguer's gut feeling, which cannot be articulated or traced to any checkable ground. Therefore X is rejected. Form B (assertion): The arguer has an unarticulated gut feeling that X. Therefore X is true.

Why it fails: An intuition offered with no traceable ground carries no more evidential weight than a guess. It neither cites the facts that would make X true or false nor explains why the arguer's hunch reliably tracks truth in this particular domain, so it cannot legitimately raise or lower the probability of X for anyone but the person having the feeling -- least of all when it is used to override evidence that has actually been produced.

• Is the intuition checkable against evidence available to others, or does it live only inside the speaker's head?
• Does the speaker's track record in this exact domain -- not an adjacent one -- justify treating the hunch as data, the way a veteran firefighter's trained 'something's wrong' differs from an untrained guess?
• Is the intuition being used to override evidence that has already been produced, or to fill a gap where no evidence yet exists?
• What support is left for the conclusion once the intuition itself is subtracted?

Legitimate when: The 'intuition' is a trained expert's fast pattern-recognition operating strictly within their narrow domain of demonstrated competence (recognition-primed judgment, e.g. an experienced clinician's or firefighter's snap read); or it is used as one defeasible input among several in a recognized philosophical method -- weighing moral or thought-experiment intuitions against theory in reflective equilibrium -- rather than as an unchallengeable trump card; or the matter is genuinely one of personal, subjective preference (whom to love, which piece of art moves you) where there is no external fact for the feeling to answer to.

Built from: Logically Fallacious · Wikipedia - Intuition (philosophy)

Appeal to Loyalty

It swaps 'prove me right' for 'prove you're on my side' — and hopes you won't notice the trade.

In the office group chat, someone points out that the new project-tracking app the team just adopted keeps losing people's data. Instead of addressing the bug, the person who picked the app replies: 'Wow. I thought you had my back on this. Guess not everyone on this team does.' Nobody ever answers whether the app actually loses data — the rest of the chat scrambles to reassure the picker that everyone does, in fact, have their back.

The tell: You notice the conversation quietly stopped being about whether the thing actually works, and started being about whether you're a good friend, teammate, or fan.

The move: Group or person G endorses, expects, or has staked its identity on position P. You are (or wish to be seen as) loyal to G. Therefore you must accept/support P — and anyone who questions P is thereby reclassified as disloyal, so the question about P itself never gets answered. Equivalently, per the structure logicallyfallacious.com gives: (1) B questions A's claim; (2) it is asserted that questioning A is itself an act of disloyalty; (3) B's claim is therefore dismissed as wrong, without engaging its content.

Why it fails: Loyalty is a fact about social allegiance — who you back, and how steadfastly. It has no bearing on whether the backed claim, policy, or person's statement is true, accurate, or well-supported. The move works by re-describing a request for evidence as a test of allegiance, so that answering the loyalty test (by falling in line) substitutes for doing the epistemic work (checking whether the claim holds up). As logicallyfallacious.com puts it, even a genuinely disloyal person's claim is not thereby false — 'A is not necessarily always right' — so the entire charge is irrelevant to the truth of the matter at issue.

• Strip out every word about loyalty, betrayal, team spirit, or 'being one of us' — is there any actual evidence for the claim left standing?
• Would a lifelong, obviously loyal member of the group get a different, gentler hearing for raising the exact same objection than an outsider would?
• Is 'disloyal,' 'traitor,' or 'not a real fan/member/employee' being used to label the act of disagreeing, rather than any actual harm done to the group?
• Notice where the conversation topic changed: did it quietly shift from 'is this claim/policy right' to 'are you loyal' without anyone announcing the switch?

Legitimate when: It is not fallacious to observe, as a separate and explicit argument, that continued group cohesion or trust has independent practical value — e.g., a team that has already committed to a plan may have good reason to execute it loyally rather than re-litigate it mid-execution, or a source's long, checkable track record of being right legitimately raises its credibility. Those are arguments about the practical cost of dissent or about track-record-based credibility, and they must be made openly and be open to challenge on their own terms — not smuggled in as an unstated demand that questioning is itself betrayal.

Built from: Wikipedia — Appeal to loyalty · Logically Fallacious — Appeal to Loyalty · Internet Encyclopedia of Philosophy — Fallacies (Group Think) · Stanford Encyclopedia of Philosophy — Fallacies

Appeal to Nature

also known as: argumentum ad naturam

Something isn't automatically good just because it's 'natural' — and it isn't automatically bad just because it's man-made.

In the group chat, a friend shares a supplement she's hyped about: 'It's literally just herbs, nothing synthetic in it — how could that possibly be bad for you?' Nobody in the thread asks what the herbs actually do once they're in your body, or mentions that nightshade and foxglove are also 'just herbs' and can stop your heart. The word 'natural' on the label is doing every bit of the convincing, and none of the actual work.

The tell: You realize the person has told you nothing about whether the thing actually works or is safe — only that it's 'natural,' as if that one word closes the case. Try naming a natural thing everyone agrees is dangerous (a hurricane, poison ivy) and watch the argument suddenly need a lot more support than it had a second ago.

The move: N is natural (or: N is unnatural). Therefore N is good/right (or: N is bad/wrong) — with an unstated bridging premise, never independently defended, that naturalness itself confers moral or practical merit (and its lack removes it).

Why it fails: The move fails on two fronts at once. First, 'natural' is descriptively slippery — it can mean 'occurs without human intervention,' 'found in the wild,' 'consistent with evolved human physiology,' or simply 'not synthesized in a lab,' and arguers switch between senses as convenient. Second, even once a sense is fixed, the step from 'is natural' to 'is good' is a non sequitur: arsenic, malaria, hemlock, and hurricanes are all thoroughly natural and bad for you, while vaccines, eyeglasses, antibiotics, and open-heart surgery are all artificial and good for you. Naturalness tracks no reliable correlation with safety, health, or moral worth on its own, so citing it supplies no actual reason for the conclusion — the premise and conclusion are related only by the positive connotation the word 'natural' carries, not by any evidence.

• Is 'natural' being used descriptively (occurs without human intervention) or normatively (therefore good), and does the argument quietly slide between the two senses?
• Would the arguer accept a natural thing they would call bad (arsenic, a shark attack, malaria) or an artificial thing they would call good (a vaccine, eyeglasses, a hip replacement)? If so, naturalness is not actually doing the justificatory work claimed.
• What independent standard — safety data, evidence of function, an explicit moral principle — is needed to bridge 'is natural' to 'is good,' and has the arguer supplied that bridge or just asserted it?
• Swap 'natural' for a neutral phrase like 'occurs without human intervention' and re-read the argument — does it still sound persuasive, or did the pull come only from the word's built-in positive glow?

Legitimate when: When 'natural' is functioning as an evidentially earned shorthand rather than doing the justificatory work itself — e.g., a specific class of natural substances has an established track record (long clinical use, known mechanism, tested safety profile) that a specific synthetic alternative lacks, and the arguer is really pointing to that track record. The move becomes fallacious the moment naturalness is treated as sufficient on its own, in place of the evidence it happens to stand in for.

Ancient: Callicles contrasts nature (physis) with human convention (nomos) and argues that nature itself shows the stronger are meant to rule and take more than the weak, treating what he takes nature to display as settling what is truly just — a claim Socrates then presses and undermines through examination. — Plato, Gorgias (the speeches of Callicles)

Built from: Wikipedia — Appeal to nature · Wikipedia — List of fallacies · Stanford Encyclopedia of Philosophy — Fallacies · Logically Fallacious — Appeal to Nature · The Fallacy Files — Appeal to Nature

Appeal to Normality

Something isn't automatically fine just because it's what everyone always does.

A new hire on the team says the constant 11pm Slack messages are wearing her out. Her manager shrugs: 'Every startup is like this, it's just how things work here.' Nobody explains why the late messages are actually necessary that night — the fact that it's the usual routine is treated as if it already answered her complaint.

The tell: You notice the answer to 'is this a good idea' was actually just 'this is what we always do' — and somehow that got treated as the same thing.

The move: X is normal, usual, or the common/standard way things are done. Therefore X is acceptable, correct, or the way things ought to be done.

Why it fails: The premise is a descriptive claim about frequency or custom; the conclusion is an evaluative or practical claim about merit or rightness. Nothing bridges the two. As the Stanford Encyclopedia of Philosophy puts it discussing this family of argument, 'that something is widely practised or believed is not compelling evidence that it is true or that it should be done.' A practice can be statistically normal and still be unsafe, unjust, or simply never examined; normality tells you what people do, not whether they should.

• Does the fact that this is normal supply any evidence of its merit, or only of its frequency?
• Would this same move have defended past widespread practices now recognized as wrong?
• Is 'normal' being used as a statistical claim (most people do this), a social norm (people expect this), or is the argument sliding between the two?
• Is the appeal to normality being used to close off an actual reason-based objection, or only to set a rebuttable default that more evidence could still overturn?

Legitimate when: Citing normality is legitimate when it is offered only as defeasible, presumptive evidence rather than as a closing move: e.g., a clinician using 'normal range' as a diagnostic baseline, an engineer citing standard practice as a starting presumption of safety subject to case-specific checks, or a court weighing customary practice as one relevant factor in a negligence standard. It stays sound only as long as it remains open to being overridden by evidence about the specific case.

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Argumentum ad populum · Wikipedia — Appeal to tradition · Wikipedia — List of fallacies · Internet Encyclopedia of Philosophy — Fallacies

Appeal to Novelty

also known as: argumentum ad novitatem

Something being newer doesn't automatically make it better — 'new' and 'improved' are not the same word.

In the group chat, someone posts: 'Throwing out my whole skincare shelf — got this serum that just dropped last week. My old stuff is basically ancient at this point, keep up.' Nobody has asked whether the new serum actually works better on her skin. The whole pitch is just that it's newer.

The tell: You notice the person hasn't said one word about results, testing, or what actually changed — just that the thing is recent. Something in you asks: newer than what, and so what?

The move: X has been around for a long time; Y is new/recent; therefore Y is better than X (equivalently: X is worse or discredited because it is old).

Why it fails: Recency and merit are independent properties unless something specific ties them together (a testing process, an iterative feedback loop, a domain where decay is real). Absent that link, citing an option's age or newness supplies no information about whether it actually solves the problem at hand — the newness is offered as if it settled the question of quality, when it only describes a date.

• Has the new option actually been tested against the incumbent on outcomes, or is 'new' the only thing being offered as evidence?
• Is this a domain with a real feedback loop that makes later versions reliably better (e.g., engineering refined through repeated testing), or one where 'new' just means 'relabeled' (fashion cycles, diet fads, rebranded products)?
• What specifically changed between the old and new version, and does that particular change bear on the merits actually at issue?
• Would the mirror-image move — 'the old one has survived longer scrutiny, so it must be better' — carry any real weight here? If not, neither move does.

Legitimate when: When the domain has a genuine mechanism connecting recency to quality — e.g., a product line refined through iterative testing and real-world feedback, a scientific finding that supersedes an earlier one because it corrected a documented error, or a case where the 'old' item has literally degraded (fresh milk vs. month-old milk). In those cases the work is done by the testing process or the physical decay, not by 'newness' as a bare label — so the justification should name that process, not just the calendar date.

Ancient: Addressing the Athenian assembly, Cleon complains that his fellow citizens are too easily won over by clever, novel arguments and the pleasure of hearing something new, and warns that steady old policies — even imperfect ones — serve the city better than constantly chasing whatever proposal sounds freshest and most dazzling. — Thucydides, History of the Peloponnesian War, Book 3 (Cleon's speech in the Mytilenean Debate)

Built from: Wikipedia — Appeal to novelty · Wikipedia — List of fallacies · Logically Fallacious — Appeal to Novelty · Wikipedia — Mytilenean Debate

Appeal to Pity

also known as: argumentum ad misericordiam, appeal to mercy

Trying to win an argument by making people feel sorry for you instead of actually answering the question they asked.

In the group chat for splitting a camping trip's costs, Jordan owes $140 because the gear left in Jordan's tent got ruined in the rain. When someone brings it up, Jordan writes: 'You guys know my dog died last week, right? I really don't need this right now.' The chat goes quiet. Nobody mentions the $140 again - even though whether the dog died has nothing to do with whether Jordan was supposed to keep the gear dry.

The tell: You notice the conversation never actually answered your question - it just made you feel bad for having asked it. The subject quietly slid from 'did this happen' to 'how could you bring this up.'

The move: Rejecting claim P (or refusing action A) would cause sympathy-evoking hardship to someone; therefore P is true, or A is owed/required.

Why it fails: Hardship and sympathy are evidence about how bad an outcome feels, not evidence about whether P is true or about whether the standard governing A has been met. The argument quietly swaps the question that is actually in dispute (did the accounts reconcile, is the defendant guilty, was the work done) for a different question (how much does this person's suffering move me), and answers that second question instead. The relevance gap is the defect: nothing about the hardship bears on the disputed fact or on the criteria the decision is supposed to track.

• Is the hardship relevant to the specific standard the decision must satisfy, or only to how bad the person (or the audience) feels about the outcome?
• Would the sympathetic facts, presented by a neutral third party with no plea attached, actually justify the conclusion under the governing standard?
• Is compassion being offered in place of evidence on the disputed point, or in addition to it?
• If the same hardship were suffered by someone whose underlying claim you already doubted, would it still feel persuasive?

Legitimate when: When the governing norm itself makes hardship, need, or mercy a legitimate factor in the decision - equitable sentencing discretion, need-based aid, humanitarian parole, a landlord choosing whether to show any leniency at all. It fails specifically when the standard in question is a factual or eligibility question (did the numbers add up, was the shot made, is the claim true) that suffering cannot answer, no matter how real that suffering is.

Ancient: Socrates observes that defendants on trial customarily bring forward their weeping children and relatives to move the jury to pity, and he declines to do so himself, arguing that a juror's oath is to judge according to law and the facts of the case, not to hand down favors to whoever seems most pitiable. — Plato, Apology

Built from: Wikipedia · Internet Encyclopedia of Philosophy - Fallacies · Stanford Encyclopedia of Philosophy - Fallacies

Appeal to Popularity

also known as: argumentum ad populum, bandwagon fallacy

A lot of people believing something doesn't make it true.

In the family group chat, someone shares a link and types, 'This detox tea has 40,000 five-star reviews, everyone's doing it, it obviously works.' Within an hour, three relatives have added it to their cart — nobody has asked what's actually in it or whether any of those 40,000 reviewers checked their blood work before and after.

The tell: You notice the person never actually explains why the thing is true or good — they just keep pointing at how many other people are already doing it, as if the headcount itself were the proof.

The move: A large number of people P believe X (or do X, or approve of X). Therefore X is true, correct, or ought to be done.

Why it fails: The size of the group holding a belief is offered as if it were evidence for the belief, but the premise (how many people hold X) and the conclusion (whether X is true) are logically independent: a belief's popularity is a fact about the psychology and sociology of a population, not a fact that was produced by, or tracks, whatever method would actually verify X. Unless the crowd's agreement was itself formed by a reliable, independent process, the number of people involved has no bearing on the truth of what they agree on, so citing that number does not raise its probability of being true.

• Why would this population reliably track truth on this particular question?
• Is popularity actually the thing at issue (e.g. a poll about opinion), or is it being substituted for a separate factual or moral question?
• Is the cited majority in a position to have reliable, independent knowledge of the disputed fact, or are they just repeating each other?
• Does the claim concern what people believe (where a count is genuine evidence) or what is true (where a count is not)?

Legitimate when: The move is legitimate in two situations: (1) when the claim under dispute is itself about the population's behavior, belief, or usage rather than about an external fact — e.g. what counts as correct grammar is set by how competent speakers actually use the language, so appealing to common usage is not fallacious but simply citing the relevant standard; and (2) when the aggregated judgment of many independent, reasonably competent assessors is a statistically reliable estimator of a shared answer (the Condorcet-style condition), so long as each person's belief was formed with some independence from the others rather than by copying the crowd.

Ancient: Socrates argues that on questions of how to live one should follow the judgment of the one who knows rather than the opinion of the many, as an athlete heeds his trainer rather than the crowd, since the many's approval does not track truth and the many 'can only do things haphazardly.' — Plato, Crito

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Argumentum ad populum · Internet Encyclopedia of Philosophy — Fallacies

Appeal to Possibility

Someone treats 'it could happen' as if it were the same thing as 'it will happen' or 'it's likely to happen.'

In the group chat planning a camping trip, one friend says, "There's a chance a bear wanders into camp, so we basically have to assume a bear is coming and cancel the trip." Nobody has seen a bear in that area in ten years. The friend isn't wrong that it's possible — but possible and expected are being quietly swapped for each other.

The tell: You notice the person never says how likely the thing actually is, or points to any real evidence for it — they just keep repeating that it "could" happen, as if that settles the matter.

The move: Let X be a claim about how things are or will turn out. The arguer establishes only the modal minimum — that X is not impossible, that X could occur, that X has not been ruled out — and then substitutes that for a claim about likelihood or about what to do: X is possible; therefore X is true, or probably true, or must be planned for as though it were. The premise fixes only that the probability of X is greater than zero; the conclusion requires some particular value for that probability, and nothing in the argument supplies one. Because the substitution happens at the modal operator rather than in the content, the same move runs equally well for not-X, which is also merely possible.

Why it fails: Possibility and probability are different quantities. 'Not impossible' only sets a floor of zero on the probability of X — it says nothing about how close to zero or one that probability actually is. The arguer swaps a claim about the logical or physical space of outcomes ('this is one of the things that could happen') for a claim about warranted degree of belief or a warranted basis for action ('this is what will happen, or what we must plan for'), without supplying the evidence that would move X from merely conceivable to actually likely.

• Has any evidence been given for how probable X is, or only that X cannot be ruled out?
• Does the argument still 'work' if you swap X for its opposite, which is presumably also merely possible?
• Is the decision at hand governed by a 'possible' standard, a 'probable' standard, or a 'proven' standard — and which one does the evidence on the table actually satisfy?
• If you grant that X is possible, does the further claim (that X is true, likely, or worth acting on) remain completely undecided?

Legitimate when: The governing standard genuinely is possibility rather than probability — e.g., a modal claim ('it is possible for X to occur') is itself the whole of what's being asserted; a precautionary or worst-case-planning context where any nonzero chance of catastrophic, irreversible harm rationally warrants a hedge regardless of exact probability (e.g., engineering safety margins); or a legal/logical context that only requires showing a possibility to defeat a claim of necessity or impossibility (e.g., rebutting 'X could not possibly have happened').

Built from: Logically Fallacious (Bo Bennett) — Appeal to Possibility · Wikipedia — Appeal to probability

Appeal to Ridicule

also known as: appeal to mockery, reductio ad ridiculum

Laughing an idea off, or making it sound silly, instead of actually saying what's wrong with it.

Someone in the group chat says, "I've been trying to eat slower and actually chew my food so I don't overeat." A friend fires back, "Oh sure, next you'll be measuring each bite with a ruler and doing a little chant before dinner 😂" and three people drop laughing emojis. Nobody ever says what's actually wrong with eating slower — the joke just makes the idea feel embarrassing to bring up again.

The tell: You laughed along, but if someone asked you afterward "okay, so what was actually wrong with what they said?" you'd come up empty. The mockery landed on a cartoon version of the idea, not the idea itself.

The move: P is restated as a mocking, exaggerated, or absurd caricature C (often a distortion of what its proponents actually hold). C is ridiculed — treated as self-evidently laughable. P is then dismissed as false or unworthy of consideration, with no independent premise ever offered that P (rather than C) is false.

Why it fails: The laughter attaches to the caricature C, not to P; unless C is shown to be an accurate and necessary rendering of P, refuting or mocking C leaves P's actual truth-conditions untouched. The move substitutes a social/emotional reaction (shared amusement, fear of looking foolish) for any premise that bears on whether P is true, so it fails on relevance even when the caricature really is absurd.

• Is the version being mocked the position its proponents actually hold, or a distorted stand-in for it?
• Strip out the jokes and the tone — what argument against the original claim, if any, is left standing?
• Does the 'that's just like believing X' comparison share the features that matter, or only a surface resemblance chosen for comic effect?
• Would the same audience still laugh if the claim were restated accurately and in its strongest form?

Legitimate when: The absurdity is earned rather than staged: a genuine reductio ad absurdum first derives, through valid steps from the claim's own commitments, a consequence that really is absurd or self-contradictory — at which point pointing out how ridiculous that derived consequence is becomes commentary on an already-established result, not a substitute for the argument.

Ancient: Socrates traces the jury's long-standing prejudice partly to comic caricature — the stage Socrates suspended in a basket, spouting nonsense about things aloft — mockery that had supplanted any real examination of what he actually taught. — Plato, Apology (recalling Aristophanes' Clouds)

Built from: Wikipedia — Appeal to ridicule · Wikipedia — List of fallacies · Logically Fallacious — Appeal to Ridicule

Appeal to Self-evident Truth

Someone dodges backing up their claim by just declaring it 'obvious' — as if disagreeing with them proves you're the unreasonable one.

In the group chat, Maya says the new roommate should pay for the whole cleaning service since 'obviously' she's the messiest one. Dev pushes back: 'Says who? Show me.' Maya types back: 'Come on, it's just obvious, I'm not going to spell it out.' She never actually points to anything — no photos, no chore log, nothing anyone in the chat had already agreed on. She's just repeating her claim in a tone that dares people to keep arguing.

The tell: You notice the person hasn't actually given you anything to check or weigh — they've just handed you their claim back wrapped in a shrug, and somehow you're the one who now looks unreasonable for asking.

The move: Claim P is the very thing in dispute. Challenged to support P, the arguer answers only that P is 'self-evident,' 'obvious,' or 'beyond question.' No independent premise, evidence, or shared starting point is offered. P is then treated as settled — not because it was shown to follow from anything, but because the arguer declared that it needs no showing.

Why it fails: The move supplies zero premises. 'This is self-evident' is not a reason for P; it is a claim about P's epistemic status that substitutes for a reason. Because the audience has not actually granted P, and P is precisely what is in dispute, calling it self-evident smuggles the conclusion in as its own warrant — a zero-premise variant of begging the question (petitio principii). It also reverses the burden of argument: instead of the arguer defending the contested claim, the listener is made to feel that doubting it is unreasonable.

• Self-evident to whom — would an informed person who actually doubts this accept it without being given a reason?
• Is the 'obvious' claim just the disputed conclusion restated in different words, rather than something distinct that supports it?
• What argument would the speaker give if pressed harder — and if one exists, why withhold it now?
• Do the people actually in this conversation already agree on the claim, or only the speaker and people who already believed it?

Legitimate when: A claim may legitimately be left undefended when it is a genuine first principle — a law of logic (non-contradiction), an immediate deliverance of perception or introspection ('I am now having an experience'), a definitional truth, or any proposition every party to the actual discussion already accepts and is not contesting. Aristotle's own test for a legitimate starting point of demonstration was that it be 'true, primary, immediate, better known than, prior to, and causative of' what it supports — i.e. not itself demonstrable without circularity, and not the thing currently in dispute. The fallacy appears exactly when that second condition fails: the claim is contested, but is asserted as if it met the first.

Ancient: Aristotle held that a proposition may serve as an undefended starting point of demonstration only if it is true, primary, immediate, better known than, prior to, and explanatory of what it supports. He warned that a debater commits a fault akin to begging the question when treating a claim as this kind of self-explanatory first principle when it is not actually one — mistaking an ordinary, disputable fact about the world for something that needs no defense. — Aristotle, on the standard for genuine first principles (as discussed in the Stanford Encyclopedia of Philosophy's entry on fallacies, drawing on the Posterior Analytics/Prior Analytics)

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Begging the question · Wikipedia — List of fallacies · Logically Fallacious — Appeal to Self-Evident Truth

Appeal to Spite

Someone talks you into agreeing not because the idea is good, but because it'll stick it to someone you don't like.

In the family group chat about splitting up Mom's belongings, someone writes: 'Let's give the piano to the community center instead of Aunt Carol — she barely even called Mom the last five years, she doesn't deserve anything nice.' Nobody has said the piano should go to whoever called most, or that the community center actually needs a piano. The whole pitch only works if you're already annoyed at Aunt Carol.

The tell: The reason on offer is really just 'wouldn't it feel good to see them lose,' wearing a costume that makes it look like it settled the question.

The move: P is urged (to be accepted, rejected, or acted on) on the ground that doing so will injure, humiliate, or thwart some person or group X toward whom the audience already feels ill will — with X's actual connection to the truth or merit of P left unstated or absent.

Why it fails: Hostility toward X is not evidence that P is true, false, wise, or unwise; it only tracks how satisfying the outcome would feel with respect to X. Swap X for a neutral or sympathetic party and the 'argument' supplies no remaining reason to accept P at all — which shows the resentment was doing all the persuasive work, not the merits.

• Does resentment toward the target bear on the truth or merits of the proposal, or only on how satisfying the outcome would feel toward that target?
• Would the argument retain any force for an audience that felt no ill will toward the target?
• If the disliked party were swapped out for a neutral or sympathetic one, does any reason to accept the claim survive?
• Is the ill will being used to do the persuading, or is it just an incidental motive sitting alongside independent evidence that was actually offered?

Legitimate when: When the audience's antipathy isn't being smuggled in as evidence but is itself the openly stated value at stake — e.g., a boycott or a principled refusal to deal with a party, where the point genuinely is 'we choose not to reward this actor's conduct,' not a disguised factual claim about something unrelated.

Built from: Wikipedia — List of fallacies · Wikipedia — Appeal to spite (redirects to Association fallacy) · Logically Fallacious — Appeal to Spite · Internet Encyclopedia of Philosophy — Fallacies · Stanford Encyclopedia of Philosophy — Fallacies

Appeal to Stupidity

Calling a real argument 'overthinking it' so nobody has to check whether the argument is actually right.

In the family group chat, Dad breaks down exactly how the new streaming bundle costs more per month than the three separate subscriptions it replaces, line by line. Aunt Rosa replies, 'Nobody needs a whole essay to know bundles save money, that's just common sense,' and three people react with a thumbs up. Nobody ever looks at the actual numbers.

The tell: You notice the facts or the math never actually get answered — they just get waved off as 'too much thinking,' and everyone treats the wave-off itself as though it settled the question.

The move: A position P is backed by careful analysis, expertise, or a chain of reasoning that takes effort to follow. The arguer treats the complexity, technicality, or expert origin of that analysis as itself disqualifying — labeling it 'overthinking,' 'egghead stuff,' or something 'plain folks already know' is wrong without it — and on that basis alone rejects P (or accepts its simple negation) without ever examining what the analysis actually says.

Why it fails: How hard an argument is to follow, or how credentialed the people who made it are, is not evidence about whether its content is true. Rejecting a conclusion because reaching it required study, or because ordinary intuition points elsewhere, substitutes an attack on the argument's source or form for an examination of its substance — the two are logically independent. An argument does not become wrong by being complicated, any more than it becomes right by being simple or widely felt; the move is a relevance failure dressed up as populist common sense.

• Is the analysis being dismissed because of a specific, identified error in it, or simply because it is complicated or came from experts?
• What is the 'common sense' verdict actually based on, if not some analysis of its own — and has that analysis been stated?
• Does calling the position 'overthinking' function as a reason it is wrong, or is it just a label standing in for one?
• Would the claim that 'plain people already know' survive being asked to give its own reasons out loud?

Legitimate when: Preferring the simple, checkable route over an elaborate analysis is not fallacious when the elaborate analysis genuinely rests on a hidden, unstated, or false assumption that a plain restatement exposes, or when direct evidence already settles the matter and the complexity is a smokescreen — provided the simplicity is actually argued for (the hidden assumption is named, the direct evidence is produced) rather than merely asserted as self-evident.

Ancient: In the comedy, the elderly farmer Strepsiades sends his son to Socrates's 'Thinkery' hoping he'll learn clever arguments to dodge his debts. The schooling instead produces a son who coolly reasons his way to justifying beating his own father. Enraged, Strepsiades leads his slaves to burn the school down, siding with plain outrage over the intellectuals' reasoning. The play never engages what, if anything, is actually wrong with the reasoning taught there; it wins the audience over by making cleverness itself look like the crime, which is the same move the fallacy trades on, applied to a whole institution rather than a single argument. — Aristophanes, The Clouds (performed 423 BCE)

Built from: Wikipedia - List of fallacies · Wikipedia - Anti-intellectualism · Wikipedia - Appeal to the stone · Wikipedia - The Clouds · Stanford Encyclopedia of Philosophy - Fallacies

Appeal to the Law

Someone acts like 'it's allowed' (or 'it's against the rules') automatically settles whether it's actually right or wrong.

A friend group chat is arguing about a landlord who jacked up rent 40% the day the old lease ended. One person types: 'He didn't break any law, so there's nothing wrong with what he did, end of discussion.' Everyone else was asking whether it was a decent thing to do to a family with two kids — not whether he'd get sued. Pointing out that it was legal didn't touch that question at all.

The tell: The conversation was about whether something is right or fair, and suddenly someone answers a completely different question — whether it's permitted or banned by some rulebook — and acts like that ends the argument.

The move: X is legal (or illegal); therefore X is morally acceptable (or morally wrong) — where legality/illegality is offered as if it settled, without further argument, a question that is actually about morality, prudence, or some other independent standard of justification.

Why it fails: Legal status and moral status are governed by distinct standards that can and do diverge. A great many prohibited acts are malum prohibitum — wrong only because some jurisdiction happened to prohibit them (parking regulations, licensing rules) — rather than malum in se — wrong in themselves regardless of what any statute says (murder, fraud). Because statutes are drafted by fallible legislators, can be unjust, can lag behind changed circumstances, and vary by jurisdiction and era, 'the law currently says so' does not by itself supply a reason that tracks moral truth. The arguer treats the legal fact as though it were logically sufficient to close the moral question, when at most it is one input a further moral argument would still need to process (e.g., a claim that law-abidingness itself has moral weight, or that this particular law reliably tracks morality) — and that further argument is exactly what's missing.

• Is the question actually at issue one of legal compliance, or one of moral/practical justification — and has the arguer quietly swapped one for the other?
• Would the arguer accept that if the statute were repealed or rewritten tomorrow, the moral verdict on the act would flip along with it? If not, legality was never really doing the justificatory work claimed for it.
• Is this a malum prohibitum case (wrong only because forbidden) or a malum in se case (independently wrong) — and is the arguer trading on the intuitive obviousness of the latter to smuggle in the former?
• Is the law being cited itself the very thing whose justice or wisdom is in dispute, making the appeal to it circular?

Legitimate when: Within a legal proceeding, adjudication, or compliance question, legality genuinely is the governing standard, so citing the law is not fallacious there — a judge ruling on whether a contract is enforceable is right to appeal to the law as dispositive. The appeal is also reasonable as one consideration among others when paired with an explicit argument that the law in question is itself well-justified (e.g., 'this safety code was written by engineers after study X, so compliance tracks safety'). It becomes fallacious specifically when legal status is substituted for moral or practical justification on a question that was never a legal question in the first place, with no such supporting argument offered.

Ancient: Creon treats the legality of his royal edict — forbidding burial of Polyneices — as settling the rightness of the act, while Antigone appeals to unwritten laws older and higher than any king's decree. The play dramatizes the gap between what a ruler has made lawful and what is actually just, though this framing as a 'fallacy' is a later analytic overlay on a literary text rather than a claim the play itself makes in those terms. — Sophocles, Antigone

Built from: Wikipedia — Appeal to the law · LogicallyFallacious.com — Appeal to the Law

Appeal to the Moon

Shooting down a decent idea by comparing it to a flawless, impossible version of itself instead of to the other real choices on the table.

In the roommates' group chat, someone proposes a chore-splitting app to replace the messy whiteboard. One roommate replies, 'Unless this app can guarantee nobody EVER forgets a chore again, what's even the point?' The app can't promise that — no app could. So the group drops the idea and keeps the whiteboard, which actually gets ignored twice as often.

The tell: The bar being demanded is something nothing could ever clear — 'never,' 'always,' '100% guaranteed,' 'forever.' The moment you notice the thing already in use couldn't pass that same test either, you know the bar was never the real issue.

The move: Option X is a real, workable option. Ideal outcome Y (a guaranteed, flawless, or permanent result) is not achievable by X, or X cannot promise Y. Therefore X should be rejected — without ever asking whether any option, including the current one, can deliver Y either.

Why it fails: The move measures a workable proposal against a standard nothing could satisfy, then treats that impossible standard as if it were the relevant bar for comparison. Because no live alternative can clear the demanded bar, the 'refutation' proves nothing about which real option is better; it only restates that reality is imperfect. The defect is a smuggled comparison class: the arguer swaps 'better than the available alternatives' for 'as good as an unattainable ideal' without saying so.

• Is the standard demanded — a guaranteed, perfect, or permanent outcome — one that any real-world alternative could actually meet?
• Does the status quo (the thing already in place) itself satisfy the standard being imposed on the new proposal?
• What non-utopian, achievable benchmark would fairly measure this proposal against its actual competitors?
• If this same 'must be flawless or it's worthless' standard were applied consistently, would it eliminate every option on the table, including the one already in use?

Legitimate when: It is not a variant of this move when the guarantee genuinely is the operative standard — e.g., a contract, safety code, or regulation that explicitly requires a guarantee, a zero-defect tolerance, or a specific certainty threshold before action is permitted. Demanding perfection is reasonable when perfection (or a stated, achievable threshold) really is what the governing rule requires; it becomes this fallacy only when the demanded standard is unattainable by any option and is used to dodge a comparison among real alternatives.

Built from: Wikipedia — Nirvana fallacy · Wikipedia — List of fallacies · Logically Fallacious — Nirvana Fallacy · Stanford Encyclopedia of Philosophy — Fallacies

Appeal to Tradition

also known as: argumentum ad antiquitatem, appeal to antiquity

Something being old, or 'the way we've always done it,' doesn't make it right.

At Thanksgiving, someone cuts the ends off the ham before baking it. Asked why, she says, 'Because my mother always did it that way, and so did her mother.' Someone finally calls grandma, who laughs: 'Oh, my pan was too small for a whole ham — that's the only reason.' Three generations followed a rule that was never about taste at all.

The tell: The only reason anyone gives is 'that's how it's always been' — nobody can say what the practice actually accomplishes, and if you ask 'but why does doing it that way actually help?' the conversation stalls.

The move: Practice, belief, or institution X has persisted since a past time T (or 'we have always done X' / 'the ancestors held X'). Therefore X is true, correct, or ought to continue.

Why it fails: The premise reports only that X has survived — that it was adopted at some point and has not been discarded since. Survival is a fact about a practice's history, transmission, and social inertia; it is not, by itself, a fact about whether X was ever tested against alternatives, still serves the purpose it once served, or is true. Something can persist purely because no one challenged it, because changing it was costly, or because the conditions that once justified it have long since disappeared. Unless the age of X is accompanied by independent evidence that it was selected for and has continued to satisfy some genuine standard, its longevity does not raise the probability that it is true or good, so citing that longevity alone does not support the conclusion.

• Does the practice's age track evidence that it works or is right, or only evidence that it has survived?
• Have the conditions that originally made the tradition sensible stayed the same, or have they changed?
• Was the tradition ever actually tested against alternatives, or has it simply never been challenged?
• Would this same argument equally have defended traditions that are now (or that the arguer already regards as) unjust or mistaken?

Legitimate when: Long, unbroken survival can be legitimate evidence, not of truth, but of practical workability, when a practice has been in continuous use under conditions similar to today's and would plausibly have been abandoned or reformed had it repeatedly failed — a Burkean 'survivorship' inference, which is weak inductive evidence about function, not proof about truth or morality, and only holds while circumstances remain relevantly unchanged. It is also not fallacious when the thing actually in dispute is the tradition or custom itself (e.g., what counts as correct etiquette, or what a community's own rule has always been) rather than an external fact the tradition is being used to settle; and purely victimless customs kept for their own sentimental or cultural value need no further justification than being cherished.

Ancient: Roman political speech routinely treated the mos maiorum — the 'way of the ancestors' — as a standard to invoke on its own authority. In his Catilinarian orations, Cicero defended taking severe action against the conspirators partly by citing earlier instances of the Senate acting forcefully against threats, framing the measures as continuous with ancestral custom rather than as unconstitutional novelty — using the practice's ancestral pedigree itself as a reason to accept it. — Cicero and Roman oratory generally (the mos maiorum)

Built from: Wikipedia — Appeal to tradition · Wikipedia — List of fallacies · Logically Fallacious — Appeal to Tradition · Internet Encyclopedia of Philosophy — Fallacies

Appeal to Trust

Someone asks you to just believe them instead of showing you the receipts.

Your landlord texts the group chat: 'I've fixed up places like this for fifteen years — trust me, that ceiling stain is just old paint, not a leak. No need to call an inspector.' He doesn't own a moisture meter, and it would cost him money if it turned out to be a leak.

The tell: You notice the person is asking you to skip the one step that would actually settle the question — and that step happens to be the one that could make them look bad.

The move: Speaker X asks audience Y to accept claim C on the strength of X's trustworthiness (relationship, reputation, tenure, or past reliability) in place of independently checkable evidence for C; therefore Y should accept C.

Why it fails: Trust in a source is at best a proxy for how reliable that source's claims tend to be — it is not itself a reason for the specific claim on the table. The proxy breaks down in exactly the cases where it is invoked hardest: when checkable evidence exists but is being withheld or waved off, when the source has a personal stake in whether C is believed, or when the source's track record covers a different kind of claim than the one now being made. The move substitutes a relationship for a reason.

• Is the request for trust accompanied by any independently checkable evidence, or offered instead of it?
• Does the speaker's track record actually cover claims of this specific kind, or only general reliability?
• Does the speaker have something to gain or lose depending on whether the claim is believed?
• Why is verification being declined right now, given that it is available?

Legitimate when: When the checkable evidence is genuinely inaccessible to the audience (time, expertise, or security constraints), the source's track record specifically covers claims of this kind, and the source discloses rather than conceals its own interest — there trust functions as an efficient, honest stand-in for verification rather than a substitute offered to avoid it.

Ancient: Aristotle treated a speaker's perceived good sense, good character, and goodwill toward the audience (ethos) as one of the three basic tools of persuasion, alongside the words themselves and the audience's feelings. Later readers of this framework caution that treating a speaker's standing as settling a claim outright, rather than as one input to weigh alongside the actual case, mistakes a stand-in for a reason. — Aristotle, Rhetoric, Book I (on ethos/pistis)

Built from: Wikipedia · Stanford Encyclopedia of Philosophy · Internet Encyclopedia of Philosophy · Logically Fallacious · Wikipedia · Wikipedia

Argument by Emotive Language

Someone picks scary or glowing words to describe a plain fact, and lets the word do the arguing instead of any actual reason.

Two roommates are splitting up chores. One texts the group chat: 'I'm not doing the dishes for someone who trashes the kitchen every night.' The other roommate left a couple of pans in the sink after cooking dinner. 'Trashes the kitchen' makes it sound like a disaster zone, so the group chat piles on before anyone actually looks at what happened — a few pans, once, after a real meal.

The tell: You notice you already feel annoyed or alarmed before you've heard a single fact you could check for yourself — the words did the convincing, not anything that actually happened.

The move: A neutral, verifiable claim P is restated using terms T that carry a strong positive or negative connotation, but no independent connotation-free content beyond P. Because T triggers an evaluative reaction, the audience accepts or rejects a conclusion C (that P is good/bad, that some action should/shouldn't follow) as though T itself were evidence for C, when T merely re-describes P without adding any reason.

Why it fails: The connotation of a word is not evidence — it is an attitude smuggled into a description. Loaded language works by what Charles Stevenson called its emotive meaning: a word can hold the same descriptive (cognitive) content as a neutral synonym while carrying a 'magnetic,' imperative pull toward approval or disapproval (Wikipedia, Loaded Language, citing Stevenson). When 'freedom fighter' is swapped for 'terrorist,' or 'investment' for 'spending,' the facts referred to do not change; only the reader's pre-loaded verdict does. The argument therefore never states, let alone defends, the evaluative premise that is doing all the persuasive work — the value judgment is presupposed by the word choice rather than argued for, which is why the Internet Encyclopedia of Philosophy classes it under the definist/loaded-language family: 'the terminology unfortunately can cause the listener to adopt those values when in fact no good reason has been given.' Wikipedia's List of Fallacies entry for the loaded label likewise treats fallacious use of an evocative term as a form of begging the question, since the term's connotation assumes the very conclusion the argument was supposed to establish.

• Strip out every emotionally loaded word and substitute the flattest, most neutral synonym you can find (e.g., 'regime' to 'government,' 'bureaucrat' to 'official') — does any actual argument survive, or does the claim go silent once the charge is drained out of the words?
• Is there a plainer, opposite-valence label for the exact same fact (a 'Boo!/Hooray!' pair, e.g. 'anti-abortion' vs. 'pro-life,' 'freedom fighter' vs. 'terrorist') — and if the speaker had used the other label, would the audience's verdict flip even though nothing about the underlying facts changed?
• Which claim is actually being defended: something about the world that could be checked, or just the feeling the chosen words are built to produce?
• Has the value judgment embedded in the word (that the thing is bad, dangerous, noble, wasteful, etc.) been independently established anywhere in the argument, or is the word the only place that judgment appears?

Legitimate when: Emotive or evaluative language is not illegitimate in itself — vivid, connotative words are a normal and often honest part of communicating a judgment one has already earned through evidence, and skilled communicators regularly use emotional framing to reinforce a case that also stands on its own reasons (Logically Fallacious, Appeal to Emotion). The move only fails when the loaded word is substituted for that supporting reasoning rather than added on top of it — i.e., when the connotation is the argument instead of a garnish on one. Calling a convicted embezzler's scheme a 'con' is fair once the deception has actually been shown; calling a disputed policy a 'con' before any deception has been demonstrated is not.

Ancient: Thucydides observes that during factional strife men changed the accepted valuations of words to suit their side — reckless daring was praised as courageous loyalty, prudent hesitation condemned as cowardice — so that the vocabulary itself delivered the partisan verdict before any argument was made. — Thucydides, History of the Peloponnesian War, Book III (on the civil strife at Corcyra)

Built from: Wikipedia — Loaded Language · Wikipedia — List of fallacies · Internet Encyclopedia of Philosophy — Fallacies · Logically Fallacious — Appeal to Emotion

Argument by Fast Talking

Someone talks fast and sounds so sure of themselves that you go along with it, even though you never actually got to check if what they said made sense.

At a group dinner, your friend says: 'So I put it on the card, but then the refund hit, minus the delivery fee, plus the tip adjustment, times the split four ways — so yeah, you actually owe me twelve more, Venmo me tonight okay next, who wants dessert.' Nobody does the math. Everyone just nods and moves to dessert.

The tell: A minute later you realize you can't repeat back a single number they said, and you already agreed to pay.

The move: Speaker asserts C while delivering the supporting steps at a pace (and with a display of confidence) that forecloses the listener's opportunity to inspect any single step. The listener, unable to locate a specific objection within the time actually available, treats that silence as assent, and treats the speed/confidence of delivery itself as though it were evidence for C. So: (fast, confident delivery of C) + (no objection voiced in the time allotted) therefore C is accepted as established.

Why it fails: Rate and manner of delivery are not evidence bearing on the truth of the content — a false claim can be said just as quickly and confidently as a true one. The tactic also exploits a time asymmetry: each assertion can be produced far faster than it can be checked (the same asymmetry that makes the related Gish gallop effective, and that Brandolini's law names directly), so 'no objection was raised' reflects insufficient time to object, not that the content withstood scrutiny. The absence of a located flaw is misread as the presence of soundness.

• Does the claim still sound convincing if it is replayed slowly, one step at a time, in writing?
• Was there an actual pause where a listener could ask 'wait, why does that follow' before the speaker moved on?
• Am I going along with this because of what was said, or because interrupting would make me look slow or rude?
• Which single step, isolated and given real time, would I actually push back on?

Legitimate when: Rapid, confident delivery is not fallacious when it summarizes points the audience has already independently verified or agreed to, when genuine time pressure (an emergency, a hard deadline) makes speed itself the correct response, or when an expert is compressing settled, checkable material for an audience that remains free to request the detail afterward and that request will actually be honored.

Ancient: Aristotle notes that a questioner can direct his line of reasoning not only against the thesis or against the answerer, but against the time itself — pressing the exchange whenever working out the correct solution would take longer than the time actually available, so that the answerer is beaten by the clock rather than by the argument. — Aristotle, Sophistical Refutations, Part 33

Built from: Logically Fallacious — Argument by Fast Talking · Wikipedia — Gish gallop · Wikipedia — List of fallacies · Internet Encyclopedia of Philosophy — Fallacies · Aristotle, Sophistical Refutations (Part 33), MIT Classics Archive

Argument by Gibberish

Someone buries you in fancy-sounding words instead of actually explaining anything, betting you'll mistake 'I don't get it' for 'they must be really smart.'

In the office group chat, someone announces a schedule change: 'We're transitioning to a dynamic bandwidth-optimized rotation framework that holistically synergizes coverage elasticity across the team.' Someone replies, 'Cool — so do I still get every other Friday off or not?' The answer that comes back is the same sentence again, slightly rearranged, with one more buzzword added. Nobody in the thread can actually say what changed, but a few people give it a thumbs-up emoji anyway because it sounded thought-through.

The tell: You nod along and feel a little behind, but if someone asked you to repeat back what was just said using your own ordinary words, you'd come up empty.

The move: Speaker asserts that X is true. Speaker 'supports' X with a passage of technical-sounding, undefined, or syntactically garbled language that does not resolve into any checkable claim. Speaker restates X. The audience treats X as established because the unintelligible middle section sounded expert.

Why it fails: The supporting material cannot function as a reason because it is never pinned to a determinate meaning — there is nothing there to affirm, deny, or connect to the conclusion by any rule of inference. What does the persuasive work instead is the listener's own confusion: people systematically rate vague, jargon-heavy statements as more profound the harder they are to parse (the 'bullshit receptivity' effect documented by Pennycook et al.), so incomprehensibility gets misread as depth or expertise rather than flagged as an absence of content.

• Can each technical term used be given a definition the speaker will commit to on the spot?
• Once paraphrased into plain language, does the passage express any claim that could turn out to be false?
• Would someone who actually works in that field recognize this vocabulary as carrying its normal, specific meaning — or is it borrowed for effect?
• If you delete the jargon, is anything left besides the conclusion restated a second time?

Legitimate when: Specialized terminology is precisely defined, used consistently with its established technical sense, and genuinely necessary because plain language would lose real content — e.g., a cited statute number, a named statistical test, or a chemical compound explained to a lay audience with each term unpackable on request. The line is whether the speaker can and will translate every phrase into something checkable; if so it is expertise, not gibberish.

Ancient: Ancient commentators said Heraclitus wrote about nature so cryptically, in riddling, oracular epigrams, that Cicero remarked he spoke 'nimis obscure' (too obscurely) and did so on purpose so as not to be easily understood; Timon of Phlius nicknamed him 'the Riddler' for the same reason. The tradition preserves this as a real complaint about mistaking deliberate obscurity for depth, not as a modern logician's example — so it is offered here as a genuine historical anchor for the broader move, not a perfect match for the argumentative trick in its current form. — Heraclitus of Ephesus, nicknamed 'ho skoteinos' / 'the Obscure' — per Cicero and Diogenes Laertius

Built from: Logically Fallacious — Argument by Gibberish · Wikipedia — Pseudo-profound bullshit · Wikipedia — Heraclitus · Wikipedia — Word salad · Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — List of fallacies

Argument by Personal Charm

Someone you like says something, and because you like them, you go along with it — even though liking them tells you nothing about whether it's true.

In a team meeting, the guy who always remembers everyone's birthday and brings donuts on Fridays says, 'Trust me, we don't need to test this update before we push it live — I've never let you down before.' Nobody asks for a test log or a rollback plan. People just nod, because he's warm and easy to like, and saying no to him feels like saying no to a friend. The update goes out untested, on the strength of good feelings toward a guy, not on any information about whether the update actually works.

The tell: You notice you're nodding along and you can't point to anything he actually showed you — no test results, no plan, nothing — just a warm feeling about him personally.

The move: Speaker S is likable, charming, warm, or personally appealing. S asserts claim P. Therefore, P is true / P deserves acceptance — where the only thing offered in P's favor is the audience's positive feeling toward S, not any evidence bearing on P itself.

Why it fails: Charm is a fact about how S makes people feel; it is not a fact about whether P is true. The argument substitutes an affective response to the speaker for an evaluation of the claim's actual support, so the stated basis for acceptance (liking S) is irrelevant to the truth of what S asserts. It works psychologically because of the halo effect — the tendency for one favorable trait (likability) to bleed into unearned trust in unrelated qualities, such as honesty or competence — but that psychological pull creates no logical connection between S being pleasant and P being so. It is the positive mirror of ad hominem: where ad hominem rejects a claim because the speaker is unlikable, this fallacy accepts a claim because the speaker is likable, and both swap in a judgment of the person for a judgment of the evidence.

• Would the claim be equally credible coming from a charmless stranger who said exactly the same words?
• Is the speaker's likability being offered at precisely the point where evidence, data, or a supporting reason was requested or expected?
• What does the case for the claim look like once the personality, warmth, and rapport are set aside — is there anything left?
• Is positive feeling toward the speaker in one respect (friendliness, humor, attractiveness) visibly leaking into trust on an unrelated respect (accuracy, competence, honesty) — the halo effect in action?

Legitimate when: Charm is not doing anything illegitimate when it merely accompanies a claim that is independently backed by evidence or genuine credentials — a well-liked expert's charm does not disqualify their expertise, it is simply irrelevant noise riding along with real support. It is also legitimate when the question actually is about the speaker's character or interpersonal effectiveness, such as hiring for a role where charisma is part of the job, or a jury weighing a witness's demeanor as one factor among others in assessing credibility. What is never legitimate is letting the charm stand in for the missing evidence on a separately verifiable claim.

Ancient: Aristotle counts the speaker's character (ethos) among the three means of persuasion, observing that hearers trust the fair-seeming speaker more readily and on contested matters almost entirely — an analysis of precisely the mechanism that becomes fallacious when perceived character substitutes for proof. — Aristotle, Rhetoric, Book I (ethos among the means of persuasion)

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies · Wikipedia — Halo effect · Wikipedia — List of fallacies · Logically Fallacious — Appeal to Flattery

Argument by Pigheadedness

Someone just plain will not change their mind -- and acts like refusing to budge is itself the winning move.

Mateo tells his roommate the dishwasher's new setting uses way less water, and pulls up the energy label, the manual, and a utility report to show it. His roommate crosses his arms and says, 'I've run it on the old setting for two years and I'm not changing now.' Mateo shows him the numbers again. Same answer: 'Not changing.' He never says the report is wrong or points to a single number that's off -- he just repeats that he won't budge, like that settles who's right.

The tell: You notice you could hand the person any proof at all and get the identical flat answer back -- so the answer was never really about what you showed them.

The move: A well-supported case for claim A is presented; person P raises no counter-evidence and no counter-argument against it, yet withholds assent purely on the grounds of not being willing to change position; P's bare non-acceptance is then treated, in the exchange, as though it were itself a reason A is false or need not be accepted -- i.e., 'A is well-argued; I object to nothing in it; therefore A is not true (for me).'

Why it fails: Refusing to be persuaded is an act, not an argument. The move quietly substitutes a psychological fact about the listener (their unwillingness to update) for a logical fact about the claim (a reason it is false), so the burden of rebuttal is never actually discharged -- it is merely declined and then treated as met. A claim's truth does not depend on whether any particular person consents to it, so persistence in disbelief, however sincere, carries zero evidential weight against the case that was made.

• Has the person named a single piece of evidence or counter-argument against the claim, or only restated that they will not be moved?
• Is there anything -- any fact, source, or demonstration -- the person would accept as grounds to change their mind, or is the position declared fixed regardless of what is shown?
• Does the person's continued refusal function in the exchange as if it were itself a rebuttal to the case just made, rather than an admission that none is available?
• When new evidence is added, is it met with a fresh, specific objection each time, or with the same blanket dismissal repeated unchanged?

Legitimate when: It is not pigheadedness merely to withhold assent to an argument one judges weak, to ask for more or better evidence before agreeing, or to decline a debate altogether -- none of these treat refusal itself as a rebuttal. The pattern only becomes fallacious once the person has been given an argument to which they can produce no actual objection, and their continued non-acceptance is nonetheless offered or treated as though it settled the question against the claim.

Ancient: After Thrasymachus insists that injustice is more profitable than justice and Socrates works through the exchange until Thrasymachus has no further objection to raise, Plato records him falling silent and, famously, blushing -- rather than conceding the point or offering a new counter-argument. Ancient and modern commentators alike have long debated whether that blush marks a real change of mind or a man who has simply run out of comebacks while still refusing, in spirit, to yield. — Plato, Republic, Book I (the Socrates-Thrasymachus exchange, ~338c-354c)

Built from: Logically Fallacious - Argument by Pigheadedness · Wikipedia - Invincible ignorance fallacy · Wikipedia - List of fallacies · Wikipedia - Thrasymachus

Argument by Repetition

Saying the same thing over and over until it starts to feel true, even though saying it again isn't a reason it's true.

At the dinner table your uncle says, "That contractor is a scammer." Ten minutes later, out of nowhere: "Seriously, total scammer." Near the end of the meal: "I'm just saying — scammer." Nobody at the table has heard a single detail about what the contractor actually did wrong, but by dessert it feels like a settled fact everyone agrees on, just because it's been said three times.

The tell: You realize you can't actually point to a reason for the claim — you just remember hearing it a lot, and that repetition alone is what's making it feel true.

The move: Claim C is asserted. C is asserted again, and again, N times, with no new evidence, reason, or argument accompanying any repetition. Either (a) the arguer treats the sheer number of assertions N as itself grounds for accepting C, or (b) the arguer treats the eventual silence or fatigue of an interlocutor who stops objecting as evidence that C has been conceded or vindicated, and C is accepted on that basis.

Why it fails: Repeating a sentence does not add a reason for it; the evidential content of 'C, C, C, ... C' (N times) is identical to the evidential content of 'C' asserted once, since no repetition contributes a premise that was not already on the table. What changes with repetition is psychological, not logical: familiarity with a claim is processed by listeners as a cue for its truth (the illusory-truth effect), and an opponent's silence can result from exhaustion, boredom, or a choice not to keep engaging rather than from being persuaded. The argument therefore substitutes a causal fact about audience fatigue or familiarity for the epistemic fact of support, so the conclusion is accepted for a reason that has no bearing on whether it is true.

• Strip away every repeated instance of the claim down to one occurrence — does any evidence or reasoning remain, or was the entire 'case' just the claim itself said many times?
• Has any new fact, reason, or argument accompanied any of the later repetitions, or is it word-for-word (or near enough) the same assertion each time?
• Is the claim's persuasiveness coming from its familiarity — from having heard it before — rather than from anything that was actually checked or shown?
• If the other side stopped objecting, is that being read as agreement or refutation, when it could just as easily be exhaustion, a change of subject, or a decision not to keep arguing?

Legitimate when: Repeating a claim is not fallacious when the repetition is doing something other than manufacturing support — for instance, restating a claim for clarity, as a reminder, as a rhetorical refrain, or alongside genuinely new evidence or argument each time (which is just repeated argument, not argument by repetition), or when repetition is used pedagogically to aid memory rather than offered as a reason to believe.

Ancient: Plutarch reports that Cato the Elder ended his Senate speeches, no matter the topic, with the declaration that Carthage must be destroyed — wearing the Senate toward war through sheer relentless repetition rather than through fresh argument on each occasion. — Plutarch, Life of Cato the Elder

Built from: Wikipedia — Argumentum ad nauseam · Wikipedia — Proof by assertion · Wikipedia — List of fallacies · Logically Fallacious — Argument by Repetition (Argumentum Ad Nauseam) · Stanford Encyclopedia of Philosophy — Fallacies

Argument by Selective Reading

Grabbing the one sentence that helps you and cutting out everything around it that would have changed what it means.

In the family group chat, Grandma writes a long text: "I could maybe come to the lake house in August, though honestly with my knee the way it is I'm not sure I could manage the stairs, so please don't plan around me." Your cousin screenshots just the words "I could maybe come to the lake house in August," crops out the rest, and posts it with "Grandma's coming, confirmed!!"

The tell: You read the whole original message and think, "wait, that's not really what they said" — every word in the quote is technically real, but the feeling of it has been flipped.

The move: A source S makes a complex statement whose overall sense M depends on several mutually qualifying parts (a claim plus its hedges, exceptions, or surrounding sentences). The arguer extracts one fragment of that statement while suppressing the qualifying parts, producing an apparent claim M' that is stronger, weaker, or otherwise different from M. The arguer then treats M' as though it were S's actual position — attacking it, endorsing it, or citing it as authority for a further conclusion C.

Why it fails: The conclusion C is only supported if S actually holds M'; but S holds M, and M' exists only because the qualifying context was cut away. The fragment is real and accurately transcribed, so the move looks like faithful citation, yet the omitted material is exactly what would determine whether the excerpt licenses C. Because the target of the argument (M') is not the target that exists in reality (M), the argument's real support is irrelevant to the conclusion it is used to reach.

• Read the quoted fragment inside its original paragraph or document — does it still say what the arguer claims it says?
• Is there a qualifier, exception, 'however,' or immediately adjacent sentence that was dropped from the excerpt?
• Would the source recognize the quoted fragment as a fair one-line summary of the position they actually hold?
• Was this particular fragment chosen because it is rhetorically convenient, rather than because it is representative of the source's overall weight of statement?

Legitimate when: Excerpting is legitimate when nothing in the surrounding material qualifies or reverses the quoted claim — i.e., a fair-minded reader of the full source would agree the excerpt states the point it is being cited for. Quoting a source's own stated bottom line, a genuine concession, or a representative summary sentence is not selective reading as long as no adjacent context changes the excerpt's force.

Ancient: In the temptation narrative, the tempter urges Jesus to jump from the height of the Temple, backing the dare with a real quotation of Psalm 91's promise that God's angels will guard and bear up the faithful. Left out is the psalm's surrounding sense — trust lived out through obedience, not tested by daring God — along with the very next line about treading safely on serpents underfoot. Jesus's reply, quoting Deuteronomy 6:16 ("you shall not put the Lord your God to the test"), is traditionally read as naming this exact move: a genuine verse invoked to license something its own surrounding sense does not support. — Gospel of Matthew 4:5–7 and Gospel of Luke 4:9–12 (the wilderness temptation of Jesus)

Built from: Wikipedia — Quoting out of context · Wikipedia — List of fallacies · Internet Encyclopedia of Philosophy — Fallacies · Stanford Encyclopedia of Philosophy — Fallacies

Argument from Age

also known as: argumentum ad antiquitatem

Something being old, or coming from someone old, doesn't make it true.

A post gets forwarded around the family group chat: 'Ancient proverb: drinking cold water right after a meal turns to sludge in your stomach.' A string of folded-hands emoji follows. Your cousin, who's a dietitian, says there's no digestive mechanism that does that, cold water just passes through like anything else. Someone replies, 'People have believed this for thousands of years — don't you think they'd know?'

The tell: Nobody answers the actual point your cousin raised. The only thing anyone offers in reply is how long people have believed it, as if the length of time something's been repeated is itself the proof, and a newer, checkable answer gets waved off for being newer.

The move: Source S (an individual elder, a previous generation, or an ancient text) is old, or has lived/existed for a long time. S asserts Y, or Y is the kind of thing older people/generations are supposed to know. Therefore Y is true, wise, or credible — because of S's age, not because of any evidence S offers for Y.

Why it fails: Chronological age or antiquity is a fact about how long a person or text has existed, not a fact about whether the beliefs that person or text holds were formed by adequate evidence or method. An elder or an ancient source may have accumulated real experience in some narrow domain, but citing raw age substitutes that unexamined proxy for actually checking the claim: it treats 'has existed a long time' as if it were 'has been tested and has survived testing.' Older individuals and older texts also predate later correctives — accumulated data, statistical method, peer scrutiny — that younger sources may have access to, so seniority in time can just as easily preserve an error as preserve a truth. The move only works rhetorically because listeners conflate being older with having earned authority, when the two are logically independent.

• Is the 'age' being invoked that of the claim or practice itself (its persistence), or the age of the person/generation/text making the claim? — the latter is this move, the former is appeal to tradition.
• Does the domain actually reward long lived, hands-on experience (e.g. a craft, a skill), or is it a domain (medicine, nutrition, technology, safety) where an older source predates the relevant evidence entirely?
• If a young person offered the exact same claim with the exact same reasons, would it be taken just as seriously — or is the age doing work the reasons aren't?
• Has anything been offered besides 'they're older' or 'it's ancient' — a track record, a mechanism, a check against reality — or does the argument stop at the source's seniority?

Legitimate when: Age can be legitimate, non-fallacious evidence in two narrow cases: first, when the property actually being claimed is itself a function of time or age (e.g., 'this wine has aged and improved,' 'this bridge is old and may have metal fatigue'), where age is directly causally relevant to the conclusion rather than a stand-in for it; second, when what's really being cited is not raw age but earned, verifiable expertise that happens to correlate with years spent practicing a skill (a craftsperson with forty years at the bench giving advice within that craft) — but in that case the age is not doing the justificatory work, the demonstrated track record is, and the argument should be stated in those terms rather than in terms of seniority alone.

Built from: Logically Fallacious — Argument from Age · Wikipedia — List of fallacies · Wikipedia — Argumentum ad antiquitatem (Appeal to tradition) · Stanford Encyclopedia of Philosophy — Fallacies

Argument from Fallacy

also known as: argumentum ad logicam, fallacy fallacy

Just because someone's reason for believing a thing is bad doesn't mean the thing itself is untrue.

Your roommate says the leftover soup is fine to eat because 'it still smells okay, and smell is the only test that matters.' You point out that smell alone doesn't reliably tell you if food has gone bad — plenty of spoiled food smells fine. Your roommate throws up their hands and says, 'Fine then, the soup is definitely spoiled!' But you never said that. You only showed that her reason for trusting the soup was shaky. The soup could still be perfectly good, or it could be bad — the smell test just can't tell you either way, and nobody has actually checked.

The tell: It feels like a clean win a beat too fast — someone pokes a hole in your reasoning and then acts like they've proven the opposite of what you said, when all they've really shown is that your reasoning didn't settle the question either way.

The move: If P, then Q. P is a fallacious argument. Therefore, Q is false. (Equivalently: 'Argument A for claim Q is fallacious; therefore Q is false.') This is a special case of denying the antecedent in which the antecedent is not a simple proposition but an entire argument.

Why it fails: Showing that an argument is fallacious only shows that the argument fails to establish its conclusion — it says nothing about whether the conclusion is actually true or false. A fallacious argument, just like a false antecedent in denying-the-antecedent, can still have a conclusion that happens to be true. Treating 'this argument for Q is bad' as though it were 'Q is false' silently swaps a claim about the quality of the reasoning for a claim about the world.

• Does the criticized argument's failure leave the conclusion merely unsupported, or does it positively disprove the conclusion?
• Are there independent grounds for the conclusion apart from the one fallacious argument that was just refuted?
• Is 'your argument for X is bad' quietly being converted into 'X is false'?
• Would the same conclusion look equally (im)plausible if a different, better argument were substituted in place of the one just knocked down?

Legitimate when: It is legitimate to conclude that the conclusion is unproven, unsupported, or not yet established by that particular argument — never that it is false. Correctly identifying a fallacy licenses withholding belief pending better evidence, not asserting the opposite of the claim.

Built from: Wikipedia — Argument from fallacy · Wikipedia — List of fallacies

Argument from False Authority

also known as: argumentum ad verecundiam

Someone with a big, impressive title says something outside the thing they're actually trained in, and people believe it anyway because of the title.

A sleep-gummy ad flashes across your phone: "Endorsed by a Nobel Prize winner!" in huge letters. You watch the clip — the man in the lab coat won his Nobel in economics, decades ago, for work on trade markets. He says the gummies "really work." He has never run a sleep study or published anything about melatonin in his life, but the word "Nobel" does all the convincing.

The tell: You catch yourself thinking "well, they must be smart enough to know" without ever checking what the person is actually smart about.

The move: Person A asserts that P is true. A is presented as an authority whose word alone should settle the question of P. But A is not a competent authority on P — A has no real expertise in P's specific domain, A's expertise lies in a different field entirely, or A is a genuine expert being cited on a question outside what they actually studied. Therefore, P is true.

Why it fails: The argument borrows the weight of expertise without the expertise actually attaching to the claim being made. A person's credentials are domain-specific: being an authority on X gives no special standing to pronounce on Y. When the cited figure's real competence lies elsewhere, the stated reason for believing P — 'because this authority said so' — is not evidence about P at all, even though it is offered and received as if it were decisive.

• What field is the cited person actually credentialed or accomplished in, and does that field match the specific claim being made?
• If you swapped in a different person with the same level of fame or credentials but in an unrelated field, would the argument feel just as convincing? If so, the expertise is doing no real work.
• Do specialists who actually work in the claim's own domain agree with what the cited authority said?
• Is the authority's name being used in place of any evidence, or is it offered alongside real, checkable support for the claim?

Legitimate when: Citing an authority is sound reasoning when the person's expertise genuinely covers the specific domain of the claim, when that view lines up with — rather than stands apart from — consensus among people who work in that domain, and when the citation supplements rather than substitutes for available evidence.

Built from: Wikipedia — Argument from authority · Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — List of fallacies · Logically Fallacious — Appeal to Authority

Argument from Hearsay

Someone tells you a friend-of-a-friend story and expects you to believe it like they saw it themselves.

At the office kitchen, a coworker says, 'Marcus told me that his sister-in-law heard from someone in HR that the whole department is getting laid off Friday.' By lunch, three more people are repeating it as fact, packing up personal photos from their desks — even though nobody in the chain actually works in HR, saw a memo, or will put their name on it.

The tell: You realize the exciting news traces back to 'someone heard from someone who heard from someone' — and every time you ask 'wait, who actually saw this happen?' the trail gets vaguer instead of clearer.

The move: Speaker S1 asserts that S2 told S1 that S2 witnessed (or was told) that P. S1 offers this chain of reported statements as though it were equivalent to direct testimony that P. Formally: 'Person 1 told me that he saw Y; therefore, I must accept that Y is true' — with the evidential weight of firsthand observation claimed for a report that in fact passed through one or more intermediate, untested tellers, none of whom is the arguer or is presented for examination.

Why it fails: Each retelling in a chain of secondhand report is an opportunity for distortion, omission, exaggeration, or outright invention, and none of the intermediate links has been checked for honesty, accurate perception, or accurate memory. The argument treats the report's arrival at the arguer's ears as if it carried the same warrant as the arguer (or a cross-examinable witness) having observed the event directly, silently discarding the compounding uncertainty introduced at every hand-off. The defect is not that hearsay is always worthless — it can shift a prior modestly — but that it is presented with a confidence appropriate to firsthand or corroborated testimony when the actual evidential chain cannot support that confidence, and the source usually cannot be located or cross-examined to test it.

• How many people stand between the original observation and the version I am hearing right now, and can each of them be named and questioned?
• Is the person telling me this an eyewitness, or are they repeating what someone else told them?
• Is there any documentary, physical, or independently corroborated evidence that does not depend on this chain of retellings?
• Would this report meet the evidentiary standard actually required for the decision being made — a serious accusation, a legal finding, a firing — or only the low bar of an interesting rumor?

Legitimate when: Secondhand testimony retains legitimate, if discounted, evidential weight when the intermediate source is known, generally reliable, has no motive to distort, and the claim is checked against independent corroboration (documents, physical evidence, or a second independent chain) — and when the confidence placed in the conclusion is scaled down to match the number and reliability of the untested links, rather than treated as settled fact.

Ancient: Socrates tells the jury he fears his 'first accusers' more than Anytus and the men formally charging him, because those first accusers spread rumors about him for years, filling the jurors' minds with tales of a Socrates who speculates about the heavens and makes the weaker argument the stronger, starting when the jurors were children and repeating it ever since. He says he cannot even name these accusers or cross-examine them — he must, in his words, 'simply fight with shadows.' The passage captures exactly the hearsay defect: an accusation absorbed secondhand over years, with no identifiable source who can be questioned about what he actually saw or heard. — Plato, Apology 18a-19a (Jowett translation)

Built from: Logically Fallacious — Argument from Hearsay · Wikipedia — Anecdotal evidence · Wikipedia — Hearsay · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies · Plato, Apology (Jowett translation), Project Gutenberg

Argument from Ignorance

also known as: argumentum ad ignorantiam, appeal to ignorance

Saying something must be true just because nobody's proven it false — or must be false just because nobody's proven it true.

Your roommate says, "Nobody's ever shown me that the new guy upstairs isn't stealing packages, so I'm pretty sure it's him." No one checked the hallway camera, no one asked the mail carrier, no one even asked the guy — there's simply no evidence either way. But your roommate talks as if that blank space where evidence should be is itself proof of guilt.

The tell: You notice the person hasn't actually shown you anything — they've just pointed at a gap where proof should be and dared you to fill it. The confidence in their voice doesn't match the amount of actual looking that got done.

The move: P has not been proven false; therefore P is true. Or, symmetrically: P has not been proven true; therefore P is false. In both directions, the arguer treats the absence of a successful proof (or disproof) of P as itself a proof of P's negation (or affirmation).

Why it fails: Failure to establish a claim is not the same event as establishing its opposite. A search that has not turned up evidence for P is consistent with at least three different situations: P is false and there was nothing to find; P is true but the inquiry was too weak, too narrow, or too recent to find it; or the matter has simply not been investigated at all. Only the first of these licenses the inference the arguer wants. Absence of evidence becomes evidence of absence only when a competent, sufficiently thorough search would have been expected to surface the evidence if it existed — and the fallacy consists precisely in skipping that check and helping oneself to the conclusion anyway, often while quietly relocating the burden of proof onto whoever is being asked to disprove the claim.

• If the claim were false, would a search this thorough have been likely to turn up evidence of its falsity — and if it were true, would this search have been likely to find that too?
• Who actually has the burden of proof here, and has the arguer quietly shifted it onto the other side by demanding a disproof instead of offering a proof?
• Is 'nobody has shown X' being treated as equivalent to 'X has been shown not to be the case'?
• Has the search for evidence been adequate, well-directed, and of the kind that would likely surface the phenomenon if it were real, or is 'no evidence' really just 'no one looked hard, or looked at all'?

Legitimate when: The inference is legitimate exactly when a competent and sufficiently thorough investigation would very probably have detected the phenomenon had it been present, so that the investigation's failure to detect it is itself informative — for example, a rigorous audit that finds no fraud after checking the relevant records, or a well-designed clinical trial that detects no side effect after adequate sample size and follow-up. In these cases absence of evidence really is evidence of absence, in proportion to how powerful the search was.

Ancient: Locke describes a debating tactic he names argumentum ad ignorantiam: pressing an opponent to either accept the arguer's proof as it stands or produce a better one of their own, and treating the opponent's failure to do the latter as though it settled the dispute in the arguer's favor — even though the opponent's silence proves nothing about whether the arguer's claim is actually true. — John Locke, An Essay Concerning Human Understanding, Book IV, Chapter XVII ("Of Reason"), 1690

Built from: Wikipedia · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies

Argument from Incredulity

Saying something must be fake or wrong just because you personally can't picture how it could work.

At the office, a coworker explains that the new noise-canceling headphones have a tiny microphone that listens to the room and plays back the exact opposite sound wave to cancel the noise out. Another coworker snorts, 'That's impossible, sound can't just erase sound, that's basically magic,' and refuses to even try them on — he's never read a word about acoustics, he just can't picture how it would work, so as far as he's concerned it doesn't.

The tell: Notice when someone's whole case is just 'that sounds made up to me' — no actual look at how it works, no checking with anyone who'd know, just a shrug and a flat 'nope, can't be.'

The move: I cannot imagine how P could be true; therefore P is false. (Or the mirror form: I cannot imagine how P could be false; therefore P is true.)

Why it fails: The speaker's failure to picture a mechanism is evidence about the limits of that speaker's knowledge, training, or imagination — not evidence about how reality is arranged. The argument quietly substitutes 'I cannot conceive of it' for 'it cannot be the case,' skipping the actual work of examining evidence, and presumes without support that the speaker's own imaginative reach is a reliable detector of what is possible.

• Is the speaker's inability to picture how it could work a fact about the world, or a fact about the speaker's own background and imagination?
• Have people who have actually studied the mechanism found it intelligible and evidenced, or is 'nobody could understand this' simply asserted?
• Would the same 'I can't picture it, so it must be false' move have rejected ideas that later proved true once they were unfamiliar (a round earth, germs, invisible radio waves)?
• Is there an actual argument against the claim's evidence on offer, or only a restated feeling of disbelief?

Legitimate when: The move is legitimate only when the inability to conceive of P rests on a demonstrated logical or physical impossibility — a proven contradiction or a settled law it would violate — rather than on mere unfamiliarity or a failure of imagination. At that point it is no longer incredulity but a genuine impossibility proof, and the burden of actually demonstrating the contradiction has been met rather than assumed.

Ancient: Augustine judged it 'too absurd' that human beings could have crossed the vast ocean to settle an antipodal landmass and there raise a race descended from the first man — and it was chiefly that sheer sense of implausibility, rather than any astronomical demonstration, that led him to doubt the antipodes were inhabited. — Augustine of Hippo, City of God (De Civitate Dei)

Built from: Wikipedia — Argument from incredulity · Wikipedia — List of fallacies · Logically Fallacious — Argument from Incredulity

Argument from Silence

Someone treats 'nobody mentioned it' as if it were proof it never happened, without first checking whether anyone actually had reason to mention it.

Two coworkers are comparing notes on a work trip a colleague took last month. One says, 'She posted eleven photos from that trip and not a single one is with her husband. They must be having problems.' Nobody stops to ask whether she ever posts photos of her husband on any trip, or whether he even came along, or whether she just doesn't like posting couple photos. The blank space in the photo album gets treated as if it were a confession.

The tell: The whole case rests on something that's missing rather than something that's there, and nobody has actually checked whether that missing thing should have shown up in the first place.

The move: A source S, which discusses matters of category C, contains no mention of X. Therefore X did not happen (or is false). The move is licensed only by an unstated bridging premise: that if X had happened, S would have mentioned it. When that bridging premise is never established, the silence is treated as if it were itself positive evidence against X.

Why it fails: Silence is genuine evidence only in proportion to how strongly the source's coverage was expected to include X — expectation set by the source's declared purpose, scope, audience, and completeness. Absent that expectation, the omission is equally well explained by countless innocent causes: the author didn't know, didn't care, had no reason to record it, was constrained by genre or space, or the record was simply lost. The fallacy substitutes 'not mentioned' for 'did not happen' without ever showing that mention was the expected default, so the silence carries none of the evidential weight the argument assigns it. John Lange's formulation makes the missing work explicit: a valid argument from silence needs (1) an identified document lacking reference to the event, (2) grounds that the author intended a complete account of that category of event, and (3) grounds that the event was significant enough the author would have noticed and recorded it. Skip any of the three and the silence is not evidence, it's just an empty space that gets filled with the conclusion the arguer already wanted.

• Would this specific source, given its purpose, scope, and intended audience, have been expected to mention X if X were true?
• Was the source attempting anything like a complete or systematic account of the category X belongs to, or only a partial, occasional, or selective one?
• Are there ordinary, non-suspicious reasons the source might have left X out even if X happened (space limits, taboo, lost text, not the author's concern, X too minor at the time)?
• Is the silence corroborated by any positive evidence elsewhere, or is it being asked to carry the whole conclusion alone?

Legitimate when: The inference is legitimate, and can be genuinely strong, when the arguer first establishes the expectation premise: the source is shown to be comprehensive or systematic with respect to the category of fact in question, the author is shown to have had the knowledge and the occasion to mention X, and X is shown to have been significant enough not to be overlooked. Historians and textual critics routinely and correctly reason this way — for example, treating Cicero's silence about certain oratorical works as good evidence they didn't exist, because Cicero was demonstrably thorough about cataloguing exactly that category of work. The same move applied to a source with no such completeness (a travel diary, a casual letter, a single eyewitness account) collapses into the fallacy.

Ancient: Some scholars have argued that because Marco Polo's account of his travels never mentions the Great Wall, tea drinking, chopsticks, or foot-binding, he could not actually have visited China and must have compiled his account from secondhand reports. Other historians reply that his book was never meant as an exhaustive survey of Chinese customs, and that several of those very features (the unified 'Great Wall' as a single monument, in particular) were not prominent or even fully in existence in the form later travelers would recognize during the period he described, so the omissions carry little weight. — Marco Polo's 'The Travels of Marco Polo', as discussed in modern scholarly debate (e.g. Frances Wood)

Built from: Wikipedia - Argument from silence · Wikipedia - List of fallacies

Argument of the Beard

Because nobody can point to the exact moment 'a little' turns into 'a lot,' someone claims there's no real difference between a little and a lot at all.

In the group chat, Maya says she's trying to stop scrolling her phone late at night. Jordan teases her: 'Okay, but what's the actual cutoff? Is midnight too late? 11:59? 11:58? You can't even name the exact minute, so there's no such thing as 'too late' — scroll all you want.' Everyone laughs, but Maya still knows that scrolling until 2am is obviously not the same as stopping at 10pm, even though nobody can point to the precise minute where 'fine' turns into 'too late.'

The tell: You notice: sure, nobody can draw the exact line — but that doesn't mean the two ends aren't obviously different. Someone is using the fuzzy middle to wave away a difference you can plainly see.

The move: Let F be a vague predicate applied along a series or spectrum n = 0, 1, ..., N, where F(0) is clearly false and F(N) is clearly true. For every adjacent pair, F(n) and F(n+1) cannot be reliably distinguished (no principled cutoff separates them). Conclusion drawn: therefore there is no real difference between F(0) and F(N), or the predicate F does not pick out anything real at all.

Why it fails: The argument treats the absence of a precise, non-arbitrary boundary as evidence that no boundary-independent difference exists. But vagueness is a feature of language and classification, not proof that the underlying property collapses. Two endpoints of a spectrum can be unmistakably different even when the middle range is genuinely unclear and no single point can be singled out as 'the' dividing line. The inference also proves too much: applied consistently, it would dissolve every graded distinction people rely on daily (hot/cold, night/day, child/adult, rich/poor), which is itself a sign the reasoning is broken rather than a discovery that those categories are illusions.

• Does the argument move from 'no exact line can be drawn' to 'no real difference exists at all'?
• Are the two endpoints of the series still obviously distinguishable, even though the middle cases are genuinely borderline?
• Would accepting this reasoning also erase distinctions everyone concedes are real, like child vs. adult or day vs. night?
• Is the vagueness being used to dodge a practical decision (a rule, a policy, a definition) that has to be set somewhere, rather than to make an honest point about definitions?

Legitimate when: It is legitimate to point out that a term is vague and that a specific proposed cutoff is arbitrary or poorly justified — for instance, objecting that a policy's exact numeric threshold was chosen without good reason. That criticism stays sound as long as it targets the placement of the line. It becomes the fallacy only when it is stretched further, into denying that the underlying distinction or category exists or matters at all.

Ancient: Ancient sources credit the Megarian philosopher Eubulides of Miletus with the sorites ('heap') and bald-man puzzles: plucking a single hair never turns a man bald, so, one hair at a time, no amount of plucking should ever produce baldness. The puzzle works because no single hair can be singled out as the one whose removal crosses the line from 'not bald' to 'bald' — yet a man with a full head of hair and a man with none are plainly different. — Eubulides of Miletus, credited in the ancient doxographical tradition as originator of the sorites and 'bald man' puzzles

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Continuum fallacy · Wikipedia — List of fallacies · Internet Encyclopedia of Philosophy — Fallacies (Line-Drawing Fallacy) · Wikipedia — Sorites paradox

Argument to Moderation

Two people disagreeing doesn't mean the truth is politely waiting in the middle.

In a group chat about whether a used car is a good deal, one friend says the transmission is definitely shot and it'll cost $3,000 to fix, another friend (who's never seen the car) says it's probably totally fine. A third friend jumps in: "You're both probably exaggerating, let's just say it needs like $1,500 in work and call it a day." Nobody has actually looked under the hood. The $1,500 number isn't based on anything except that it sits neatly between the other two guesses.

The tell: You realize the 'reasonable' number or position didn't come from anyone actually checking the facts — it came from averaging two people's opinions, one of whom might just be totally wrong.

The move: A asserts P; B asserts not-P (or some Q incompatible with P); therefore the truth (or the correct policy) lies at the midpoint between P and Q. More generally: two positions are staked out on a scale; the mere existence of that spread, and nothing else, is treated as evidence that the correct answer sits between them.

Why it fails: Where the truth lies on a question is fixed by the evidence bearing on that question, not by where the disputants happen to have planted their flags. The distance between two claims carries no information about which claim (if either) is closer to correct — the midpoint is just as easily an average of one true and one false claim, two false claims, or two claims that were never on a single continuous scale to begin with (a false-dilemma error hiding inside the setup). Treating 'the disagreement exists' as itself evidence for a compromise position also creates a perverse incentive: a party can drag the apparent 'moderate' answer toward its own view simply by staking out a more extreme opening position (anchoring), which shows the midpoint is being generated by rhetorical positioning, not by evidence.

• Is the midpoint being offered as evidence of what's true, or is it being defended as a plain compromise between competing interests (which can be perfectly legitimate)?
• Would the 'reasonable middle' shift if one side had simply opened with a more extreme claim — and if so, is anchoring doing the work instead of evidence?
• Grant one of the two claims outright: does the case for the midpoint conclusion survive, or does it evaporate because it was never anchored to anything but 'both sides said something'?
• Is there independent evidence — measurement, expert consensus, a fact of the matter — that one endpoint is simply right and the other simply wrong, making 'splitting the difference' actively worse than picking a side?

Legitimate when: Splitting the difference is a sound and often necessary move when the question is not 'what is true' but 'what should we jointly do' — negotiating a price, a settlement, a deadline, or any situation where both parties' stated interests genuinely deserve partial weight and there is no independent fact of the matter to discover. It is also reasonable when the two extremes really do bracket an unknown continuous quantity and no better estimate exists, so the midpoint is the least-bad guess pending more evidence — but that has to be argued for on its merits, not simply assumed because two people disagree.

Built from: Wikipedia · Wikipedia — List of fallacies

Argument to the Purse

also known as: argumentum ad crumenam, appeal to wealth

Someone assumes a person must be right just because they have a lot of money -- or must be wrong because they don't have much.

At a family dinner, someone brings up a diet trend. Uncle Ray cuts in: 'Cousin Sasha built the most profitable landscaping company in three counties. When Sasha says cutting out carbs is what works, that's who you listen to -- not some magazine article written by someone who's never made real money.' Nobody at the table actually looks into whether cutting carbs does what Sasha claims. They just nod, because Sasha is the one who's clearly done well for himself.

The tell: You notice the conversation quietly swapped 'is this actually true' for 'has this person made a lot of money' -- and somehow everyone acted like the second question answered the first, even though running a landscaping company and understanding nutrition have nothing to do with each other.

The move: Speaker S is wealthy (or, in the mirror form, poor). S asserts claim P, where P's truth does not depend on S's net worth. Therefore P is true (or, in the mirror form, false).

Why it fails: Net worth is evidence about how much money someone has accumulated, not evidence about the world that most claims describe. Unless the specific claim P is itself a claim about the very skill that produced the wealth (e.g., 'this pricing strategy grows revenue,' asserted by someone who demonstrably grew revenue with it), the size of S's bank account carries no logical weight toward P. The argument substitutes an easily observed, emotionally loaded fact (rich or poor) for the actual evidence P would require, and it typically supplies no independent reason to accept or reject P at all — wealth is doing all the work that argument should be doing. The fallacy is symmetric: the negative form (argumentum ad lazarum) discredits a claim because the speaker is poor, on the equally unsupported assumption that poverty tracks bad judgment.

• Would the claim's plausibility change at all if the speaker's bank balance were different, everything else held constant?
• Is wealth actually correlated with expertise on this specific question (e.g., a claim about running a particular kind of business), or is it functioning as a general-purpose halo applied to an unrelated topic?
• Is the mirror-image move ('the poor speaker is therefore wrong, or therefore uncorrupted and right') equally tempting here? If either direction feels persuasive, wealth itself is not doing any evidential work.
• State the trait invoked (rich/poor) and the claim being judged separately — is there a stated bridge connecting the two, or does the argument jump straight from net worth to verdict?

Legitimate when: A speaker's financial success is legitimately relevant only when it is itself the direct, verifiable track record of the exact skill the claim is about — for instance, a specific investor's realized returns bear on a claim about that investor's specific strategy, or a contractor's history of profitable, well-reviewed jobs bears on a claim about their bidding practices. Even then, the wealth is a proxy for a documented track record, not a free-standing credential; citing 'they're rich' without showing that the riches came from the relevant skill, applied to the relevant kind of problem, reintroduces the fallacy.

Built from: Wikipedia — Argumentum ad crumenam · Wikipedia — List of fallacies · Internet Encyclopedia of Philosophy — Fallacies (Appeal to Money / Ad Crumenum)

Arguments

This isn't actually a trick anyone uses to win a chat or a debate; it's a stray title card that wandered into the collection by mistake.

Picture a recipe box where every card is a dish, except one card near the back just says 'Recipes' in someone's handwriting. It isn't a dish. It's the label that was supposed to sit on the outside of the box, not inside it with the actual recipes. This entry, titled only 'Arguments,' is that stray label: it looks like it belongs on the shelf with the real tricks, but open it up and there's no trick inside, just the word that used to head the list.

The tell: There's no gut-check moment here, because there's no scene to feel uneasy about. If something is labeled 'Arguments' and offered up as if it names one specific sneaky move, the odd feeling is simply confusion: which move, exactly?

The move: There is no schematic form to give here, and that absence is itself the finding. A record's logical form names the moving parts — a statement S, an arguer A, a class F — and shows what gets substituted for what; no such substitution can be written for this entry, because the bare label 'Arguments' picks out no movement from anything to anything. What the label appears to name is a container rather than a move: the category heading under which real fallacies were filed in whatever source inventory this row was harvested from. A container has no failure condition, and the honest content of this field is the statement that the field cannot be filled.

Why it fails: There is no argumentative defect to analyze here. Cross-checking the Stanford Encyclopedia of Philosophy entry on fallacies, the Wikipedia 'List of fallacies,' and the Internet Encyclopedia of Philosophy entry on fallacy confirms that none recognizes a distinct fallacy simply named 'Arguments.' The record's own seed audit already flags it as 'an index or methodological item rather than a substantive fallacy,' almost certainly a scraping artifact carried over from a source inventory's section heading (e.g., a page titled 'Arguments' that introduces a list of actual fallacies). Treating it as a fallacy in its own right would misrepresent the taxonomy rather than describe a real reasoning error.

• Is 'Arguments' the name of a specific inferential move, or is it a section/category heading from a source list?
• Can this label be pointed at one concrete pattern of bad reasoning, the way 'begging the question' or 'appeal to authority' can?
• Does any of the three cross-checked reference works (SEP, Wikipedia, IEP) list 'Arguments' as a standalone fallacy entry?
• If retained in the inventory at all, should this record be reclassified as a non-fallacy reference/index node rather than searchable alongside genuine fallacies?

Legitimate when: Not applicable. There is no argumentative technique named by this record to be legitimate or illegitimate depending on context; the record itself needs reclassification, not a context in which the 'move' is sound.

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — List of fallacies · Internet Encyclopedia of Philosophy — Fallacies

Avoiding the Issue

Getting asked one thing and answering a totally different thing, while acting like you actually answered it.

You lent a friend $40 three weeks ago. You text: "Hey, can you send that $40 back?" They reply: "Work has been absolutely insane lately, I'm barely sleeping, and honestly I noticed you posted vacation photos last week -- must be nice to have spare cash." Three paragraphs later, the $40 has never been mentioned again.

The tell: You suddenly find yourself defending or explaining something that has nothing to do with what you brought up, and the thing you actually asked about has quietly vanished from the conversation.

The move: A challenger poses question or claim Q1 (the issue actually at stake). The responder constructs an argument -- possibly internally cogent -- that bears on a distinct proposition Q2, and then proceeds as though Q1 has thereby been addressed, discharged, or settled.

Why it fails: The internal soundness of the argument for Q2 is irrelevant to its dialectical job: no logical or evidential bridge is ever built connecting Q2 back to Q1, so Q1's truth-value and its burden of proof are left exactly where they started. The defect is not that the reasoning offered is bad reasoning -- it is often perfectly good reasoning -- but that it is good reasoning aimed at the wrong target, which earns no license to treat the original point as answered.

• Write down, in one sentence, the exact question or claim that was actually put to the responder.
• Grant everything the response asserts as true -- does the original question still stand completely open afterward?
• Is the topic the response lands on easier to defend, more flattering, or more emotionally loaded than the original topic?
• Would a neutral moderator, on hearing this answer, have to say 'that isn't what was asked' and repeat the original question?

Legitimate when: When both parties have explicitly agreed to narrow or reorder the question (e.g. 'let's first settle whether X is even possible before arguing about whether it happened'), or when the second issue is a genuine logical precondition that must be resolved before the first can be meaningfully addressed -- that is legitimate scope management, not evasion.

Ancient: Aristotle describes refutations that go wrong through ignorance of what a genuine refutation requires: to refute a claim you must establish its actual contradictory, and a disputant who instead proves some other conclusion -- one merely likely to be mistaken for the real target -- has not refuted anything, however sound that other proof may be on its own terms. — Aristotle, On Sophistical Refutations

Built from: Wikipedia - Irrelevant Conclusion (Ignoratio Elenchi) · Wikipedia - Red Herring · Stanford Encyclopedia of Philosophy - Fallacies · Logically Fallacious - Red Herring

Base Rate Fallacy

also known as: base rate neglect

Getting swept up by one striking clue and forgetting to ask how common the thing it points to actually is.

Your friend's smartwatch buzzes: 'Irregular heartbeat detected — 99% accurate.' She's convinced she has a heart problem and starts researching cardiologists. Nobody at the table stops to ask how many healthy 28-year-olds get that same buzz on a false alarm, or how rare heart arrhythmias actually are in someone her age. The '99%' number does all the talking.

The tell: A big, confident percentage gets repeated like it already settles things, and nobody in the conversation has asked 'okay, but out of everyone who'd get this same result, how many actually have it?'

The move: From a strong case-specific indicator — e.g. P(Evidence | Condition) is high, such as a test's reported accuracy — the arguer infers P(Condition | Evidence) is high, without incorporating P(Condition), the base rate (prior prevalence) of the condition in the relevant population.

Why it fails: This silently drops a required term of Bayes's theorem: P(Condition | Evidence) = [P(Evidence | Condition) x P(Condition)] / P(Evidence). When P(Condition) is small, P(Evidence) is dominated by false positives drawn from the much larger pool of non-cases, so the true posterior probability can be far lower than the cited accuracy figure implies, even though every individual number offered along the way was correct. The error is not a false premise but an incomplete update.

• What is the base rate of the condition in the actual population being reasoned about, and was it ever stated out loud?
• Rebuilt as a natural-frequency table (e.g., per 10,000 cases), how many of the 'positive' results turn out to be true cases versus false alarms?
• What reference class supplies that base rate, and would picking a different, equally defensible reference class change the answer?
• Is the impressive-sounding percentage the test's accuracy (its sensitivity or specificity), or the actual probability of the condition given this particular result — and is the arguer quietly treating those two different numbers as though they were the same?

Legitimate when: When the base rate is genuinely uninformative (close to 50/50) and so adds nothing to the calculation, when it has already been folded into the probability being cited, or when the likelihood ratio is so extreme that it overwhelms even a very small prior (e.g., a DNA match specific to one in several billion people).

Built from: Wikipedia — Base rate fallacy · Wikipedia — List of fallacies

Begging the Question

also known as: petitio principii, assuming the conclusion

It tries to prove something true by just saying the same thing again in different words, as if that counted as a reason.

In the group chat, someone posts: "This is hands-down the best taco place in the city." A friend asks, "How do you know?" The reply: "Because it's obviously the top spot for tacos around here." Nobody actually said anything new — "best taco place" and "top spot for tacos" are the same claim, just reworded, so the question "how do you know" never actually got answered.

The tell: You notice you can't repeat the reason back in your own words without it turning into the exact same claim you were questioning. It feels like the conversation went in a circle and you ended up back where you started, no closer to being convinced.

The move: A conclusion C is defended by a premise P, but P is not independent support for C — either P and C are the same claim in different wording, or accepting P already requires having accepted C. The chain of support therefore runs from C back to C rather than from external grounds to C.

Why it fails: A proof is supposed to give a doubter of C a reason to accept C that does not itself depend on already accepting C. Here it cannot: anyone who denies the conclusion has exactly the same reason to deny the premise, since the premise just is the conclusion wearing different words (or is answerable only by presupposing the conclusion). The argument may even be formally valid — C does follow from P — but it makes no epistemic progress, because the 'support' is not independent of what it is supposed to support. Stanford Encyclopedia of Philosophy notes this is why the defect is best classed as epistemic rather than strictly logical: the argument's weakness lies in what counts as a reason, not in whether the conclusion follows.

• Could someone who genuinely doubts the conclusion accept the premise on its own merits — or does accepting the premise already require having accepted the conclusion?
• Rewrite each premise in completely independent vocabulary: does any premise collapse into a paraphrase, synonym, or intensified restatement of the conclusion?
• Trace the chain of support from the conclusion backward through each reason offered — does it eventually loop back to a claim it started from?
• Strip out the rhetorical padding ('obviously,' 'clearly,' 'at its core'): is the 'evidence' actually just the claim said twice?

Legitimate when: Circularity is legitimate when it functions as coherence-building rather than as proof to a doubter — e.g., a dictionary's large web of terms that mutually define one another, or a scientific theory whose parts mutually support each other within an already-accepted framework. Internet Encyclopedia of Philosophy makes this distinction explicit: a small, tight circle (A because B, B because A) is fallacious, but a large circle spanning many independent claims can be informative and is not treated as fallacious, because it is not being used to force assent from someone who rejects the whole system. It is also legitimate to state something as a premise and its logical restatement as a working definition, so long as no one is being asked to treat the restatement as newly-won proof.

Ancient: Aristotle treats 'asking for the original point' (petitio principii) as a standing flaw of dialectical argument: in a debate the answerer must not be made to grant, even in disguised wording, the very thesis that the argument was supposed to establish, since doing so lets the arguer 'obtain' the conclusion rather than prove it. — Aristotle, Topics and Prior Analytics (with parallel treatment in On Sophistical Refutations)

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Begging the question · Internet Encyclopedia of Philosophy — Fallacies

Biased Sample Fallacy

Counting only the people who showed up, and then talking as if you counted everybody.

A gym posts a sign: "94% of members who took our 6am bootcamp say it changed their life — ask anyone in the class!" Sure — the only people in that room at 6am are the ones who loved it enough to keep showing up. Everyone who tried it once, hated it, and quit isn't standing there to be asked.

The tell: The number sounds huge and confident, but when you ask "who exactly did they ask, and how did those people end up in the room?" you realize the room was already stacked with the answer they wanted before anyone said a word.

The move: A proportion Q of sample S has attribute A, and S was drawn by a mechanism in which membership in S is correlated with A; therefore proportion Q of population P (which S was meant to represent) has attribute A.

Why it fails: The inference from sample to population silently assumes the sample was drawn so that every member of the population had a fair (equal or known) chance of inclusion. When inclusion in the sample is itself correlated with the very outcome being measured, the sample statistic is not an estimate of the population statistic at all — it estimates the statistic for 'the kind of person/case likely to end up in this sample,' a different and usually unstated population. Enlarging the sample does not fix this: a bigger sample built by the same selective mechanism just reproduces the same skew with tighter (and therefore more misleadingly confident-looking) error bars. This is why the defect is distinct from hasty generalization, which is a size problem (too few cases) rather than a mechanism problem (the wrong cases).

• Who or what could not enter the sample, and does that exclusion track the outcome being measured?
• Would collecting more data through the same channel fix the problem, or would it just produce a larger version of the same skew?
• Did people/cases enter the sample by self-selecting (volunteering, showing up, responding), and is willingness to self-select plausibly linked to having the measured trait?
• What does the sampling frame (the list or venue the sample was drawn from) actually cover, compared to the population the conclusion is stated about?

Legitimate when: The generalization is legitimate when the sample is drawn so that inclusion is independent of the measured attribute — e.g., true random sampling, stratified sampling that corrects known population proportions, or a census-like method where nonresponse/exclusion is demonstrably unrelated to the outcome. In that case, projecting the sample frequency to the population is sound (subject to ordinary margin-of-error caveats).

Built from: Wikipedia - Sampling bias · Wikipedia - List of fallacies · Wikipedia - Faulty generalization · Wikipedia - Cherry picking (fallacy) · Internet Encyclopedia of Philosophy - Fallacies

Blind Authority Fallacy

Someone says "because so-and-so said so, end of discussion" — and you're expected to just believe it and stop asking questions.

In the family group chat, your uncle posts a screenshot: "My trainer says cold showers cure inflammation, that's literally his job, I'm not going to argue biology with a fitness professional." Your cousin gently asks if there's a study behind that. Uncle replies: "Bro has a certification. What certification do you have? Drop it."

The tell: You notice nobody's actually answering the question — the second you ask "how do we know that's true," the person gets annoyed or acts like asking was the rude part, instead of just pointing to evidence.

The move: Authority A asserts that P. A's assertion is treated as final and unquestionable — not because A's competence and track record were checked, but because A is not to be questioned. Therefore P, and any request to check P or A is itself treated as illegitimate.

Why it fails: An appeal to authority can be a reasonable, if defeasible, form of inductive support when the authority actually has relevant expertise and the claim remains checkable in principle. The blind-authority move strips out both safeguards: it does not establish that A is competent in the specific domain of P, and it forecloses in advance any procedure — evidence, cross-examination, disagreement among other experts — by which P could be revised. The authority's say-so is made to substitute for the reasons that would normally transmit evidence for P, rather than serving as one input to be weighed alongside them; contrary evidence is not rebutted, it is ruled out of order.

• Is the authority speaking within a domain where they have actually demonstrated competence, or merely a domain adjacent to their prestige?
• Is there any procedure, even in principle, by which the authority's claim could be checked, tested, or overturned?
• Does the arguer treat the act of questioning the authority as itself illegitimate, rude, or out of bounds?
• If other qualified authorities disagree, is that disagreement acknowledged and weighed, or simply not mentioned?

Legitimate when: Citing an authority is legitimate when the authority has genuine, verifiable expertise in the specific matter at hand, the claim remains open to being checked against evidence, and disagreement among comparably qualified authorities is acknowledged rather than suppressed. In that case the citation functions as defeasible evidence pointing toward a conclusion, not as a conversation-ending substitute for it.

Ancient: Cicero complains of disputants who rest their whole case on the master's bare say-so: the Pythagoreans, he reports, would answer any challenge with "he himself said it" (ipse dixit), letting Pythagoras's authority stand in for an actual reason — a habit Cicero treats as an enemy of real inquiry. — Cicero, De Natura Deorum, Book I

Built from: Wikipedia — Argument from authority · Wikipedia — List of fallacies (ipse dixit) · Internet Encyclopedia of Philosophy — Fallacies · Stanford Encyclopedia of Philosophy — Fallacies

Broken Window Fallacy

also known as: broken window parable

Saying that breaking something is secretly good for the economy because somebody gets paid to fix it.

Your roommate knocks your phone off the counter and the screen shatters. In the group chat someone jokes, 'Well hey, you're basically supporting the local phone-repair guy — good for the economy!' Everyone laughs, and the conversation moves on like the broken phone was practically a nice thing to have happen.

The tell: The repair money has to come from somewhere, and wherever it was going before — new shoes, a concert ticket, savings — just quietly disappeared from the story, and nobody at the table mentioned it.

The move: An event D destroys or diverts a resource R and forces replacement spending V (e.g., hiring a glazier, a repair crew, a rebuild). The visible transaction V is then cited as proof D was economically beneficial, while the unseen counterfactual is dropped: had D not occurred, R (or the money now spent replacing R) would have funded some other purchase or investment V'. The argument counts V as a net gain instead of a mere transfer that replaces V' and the lost resource R itself, so it mistakes gross activity for net wealth.

Why it fails: The inference conflates 'money changed hands' with 'value was created.' It credits the seen effect (the glazier's new business, the contractor's new contract) while never pricing in the unseen effect (the shopkeeper's foregone shoes, the insurer's or taxpayer's foregone alternative spending, and the plain fact that society now has only a window/building instead of a window/building plus whatever else the money would have bought). Because the counterfactual spending is invisible and diffuse — spread across many sellers who never notice the sale that didn't happen — it is easy to omit from the ledger while the beneficiary of the repair spending is concrete and easy to point to. The result is a causal claim (destruction -> net economic benefit) that survives only because half the causal picture, the opportunity cost, was never examined.

• What would the destroyed or diverted resources have paid for if the destructive event had not happened?
• Is 'money changed hands' or 'people were hired' being treated as proof that new wealth was created, rather than as a transfer that replaces a loss?
• Who is the invisible loser here — the merchant, taxpayer, or investor whose foregone purchase or foregone return never shows up in anyone's count?
• Does the same logic still sound right if you scale the destruction up — would breaking every window in town, or burning every warehouse in the district, produce an even bigger 'benefit'? If the scaled-up version is absurd, the small version was never a real benefit either.

Legitimate when: The 'rebuild is good' half of the claim is not automatically wrong — it becomes sound reasoning, not this fallacy, when the resources being put to work were genuinely idle rather than diverted from another use (e.g., unemployed labor and idle capital in a deep slack economy that the repair spending activates with no real opportunity cost), or when the replacement is a net upgrade rather than a like-for-like repair (e.g., destroyed infrastructure is rebuilt to a higher standard than would otherwise have been funded, so society ends up with more than it had, not merely the same thing twice-paid-for). The move is legitimate only when someone can actually show the resources had no better foregone use — asserting this is not enough; it must be demonstrated.

Ancient: A boy breaks a shopkeeper's window pane. Onlookers console the shopkeeper by noting the glazier now has six francs of business he wouldn't otherwise have had, and reason that broken windows keep glaziers employed and thus benefit trade. Bastiat replies that this reasoning stays 'confined to that which is seen; it takes no account of that which is not seen': had the window survived, the shopkeeper would have spent those six francs on something else, say new shoes, and society would have ended up with both a window and a pair of shoes rather than a window alone. The glazier's gain is exactly offset by the shoemaker's loss; destruction has not created a franc of new wealth, only relocated existing spending while destroying a pane of glass. — Frédéric Bastiat, "That Which Is Seen, and That Which Is Not Seen" (Ce qu'on voit et ce qu'on ne voit pas, 1850)

Built from: Wikipedia — Parable of the broken window · LogicallyFallacious.com — Broken Window Fallacy

Bulverism

also known as: psychogenetic dismissal

Instead of proving someone wrong, you just explain why they'd say something like that.

In the group chat, Priya posts a link showing the new apartment complex will raise everyone's parking costs. Dev replies, "Of course you're against it, you already have a reserved spot," and the chat moves on, nobody checks the numbers in Priya's link, and nobody notices that Dev never said the numbers were wrong.

The tell: You notice the other person talked entirely about you, your reasons, your background, or your track record, and never once touched the actual claim, the facts, or the evidence you brought.

The move: Person A asserts P. Person B, without refuting P, explains why A believes P by citing A's psychology, motive, identity, or circumstances M. B then treats this explanation of A's belief as though it settled the truth of P (usually as grounds to dismiss P).

Why it fails: The causal or psychological origin of a belief is irrelevant to whether the belief is true. Even a badly-motivated, biased, or self-interested person can be right; even a neutral, well-motivated person can be wrong. Bulverism substitutes an account of how someone came to hold a claim for an account of whether the claim holds up, and treats the first as if it answered the second. It compounds two moves: it assumes, without argument, that the claim is false (a circularity), and it explains that assumed falsehood by an irrelevant genetic story about the arguer (the genetic fallacy / circumstantial ad hominem), so the claim's actual content and evidence are never engaged.

• Has the arguer ever actually shown the claim is false, or only explained why the person supposedly believes it?
• Would the psychological or circumstantial story offered explain the belief just as well if the claim were true?
• Could the exact same move be turned around and used to dismiss the arguer's own belief on the arguer's own motives or background?
• Strip out every sentence about the speaker's identity, motive, or psychology, what evidence against the actual claim is left?

Legitimate when: It is legitimate to raise a person's circumstances or motive as a reason for extra scrutiny of their evidence, or to explain a documented error after the claim has independently been shown false, or to diagnose bias as a supplement once the argument itself has already been rebutted on the merits. It becomes Bulverism only when the motive-story replaces engagement with the claim rather than following independent refutation of it.

Built from: Wikipedia — Bulverism · Wikipedia — List of fallacies · Logically Fallacious — Bulverism · Stanford Encyclopedia of Philosophy — Fallacies

Causal Reductionism

Picking one real ingredient out of a recipe and pretending it's the whole dish.

In the group chat about why the softball team's season fell apart, someone posts: 'It's simple -- Coach benched our best hitter in June, that's the whole story, bring her back and we win state.' Nobody mentions that the team also lost its catcher to a transfer, switched practice fields to a lot with no lights, and played six of its last eight games on the road. The bench decision was real and it did hurt. It just wasn't the only thing going on.

The tell: The explanation feels satisfyingly tidy -- one villain, one fix, one clean story -- for a mess that everyone around the table knows had several moving parts.

The move: C contributed to E, where C, A, B, ... were jointly sufficient to produce E; therefore C alone caused E (and C is 'the' cause, 'the' explanation, or the sole lever for changing E).

Why it fails: The argument treats a set of jointly-operating, conjoint causal factors ('A and B and C') as if they were mutually exclusive rival candidates ('A or B or C'), then declares one winner. This is a form of false dilemma applied to causation: the premises establish that C is A cause, but the conclusion smuggles in that C is THE cause, discarding the other factors that were actually needed alongside it to produce the effect. Nothing in the evidence rules out the other contributors, so the move from 'a cause' to 'the cause' is unsupported.

• Is the cited factor claimed to be the sole cause, or is it really just one contributor among several that operated together?
• What other conditions were present and jointly necessary for the effect to occur -- would the effect still have happened without them, even with C present?
• What share of the effect remains unexplained after the cited cause is fully accounted for?
• Would intervening on this single factor alone plausibly remove or reverse the effect, as the argument's proposed remedy implies?

Legitimate when: When evidence (an experiment, a controlled comparison, or an established mechanism with dose-response data) actually shows the cited factor is independently sufficient -- or overwhelmingly dominant -- and that the other candidate factors are negligible or held constant. Isolating a genuine root cause among true contributing conditions, with the other conditions ruled out or shown not to matter, is legitimate causal reasoning, not the fallacy.

Built from: Wikipedia - Fallacy of the single cause · Wikipedia - List of fallacies · Internet Encyclopedia of Philosophy - Fallacies · Stanford Encyclopedia of Philosophy - Fallacies

Cherry Picking

also known as: suppressed evidence, fallacy of incomplete evidence

Showing only the wins and hiding the losses so the record looks better than it really is.

A boutique gym posts a carousel: three clients, dramatic 8-week transformations, huge smiles. What the post doesn't say: forty-two other people started that same 8-week program. Thirty saw no visible change, nine quit early, three actually gained weight. The three photos are real. The picture they paint is not.

The tell: You notice the person only ever shows success stories, never a single case that went badly, and when you ask 'okay, but what about the people it didn't work for,' they get vague or change the subject.

The move: From evidence set E = {e₁, e₂, ..., eₙ}, arguer presents only the subset Eₛ ⊂ E whose members support claim P, while relevant members of E \ Eₛ that weaken or contradict P are available, known (or knowable) to the arguer, and withheld. The audience is invited to treat Eₛ as if it were representative of E.

Why it fails: The defect is not in any single cited fact — each cherry-picked data point can be entirely true. The defect is that the selection rule used to build the evidence set was chosen because of its outcome (it agrees with P), not because of any independent criterion of relevance or quality. This breaks the link between 'the evidence I was shown' and 'the evidence that exists,' so the sample can no longer license a general or comparative claim about the whole body of evidence. It is a species of weak induction: the inference from Eₛ to P is only as strong as Eₛ is representative of E, and a motivated selection rule guarantees it isn't.

• What was the rule for including or excluding evidence, and would that same rule have been used before anyone looked at the results?
• What contrary or countervailing evidence existed in the same dataset, time period, or population, and why isn't it shown?
• If the identical selection rule were applied to the full evidence set rather than the arguer's chosen slice, would the conclusion survive, weaken, or reverse?
• Is the withheld evidence roughly as strong, or stronger, in relevance and reliability as the evidence that was shown?

Legitimate when: Selecting a subset of evidence is legitimate — not fallacious — when the selection criterion is relevance- or quality-based, is fixed independently of whether it helps or hurts the arguer's position, and is disclosed. A meta-analysis that excludes studies for a stated, outcome-blind methodological flaw; a lawyer citing only the controlling precedent because non-controlling cases are legally irrelevant; a doctor ordering the specific tests indicated by a patient's symptoms rather than every test that exists — none of these is cherry picking, because the filter is applied the same way regardless of which result it would produce.

Ancient: Shown to a wall of votive paintings at a temple on Samothrace, each depicting a sailor who prayed to the gods and survived a shipwreck, a friend asked Diagoras — known as 'the Atheist' — whether the pictures didn't prove the gods answer prayer. Diagoras replied that of course they did, since nowhere were there any paintings of the sailors who prayed just as hard and drowned anyway. — Cicero, De Natura Deorum, Book III (via the character Cotta, recounting an anecdote about Diagoras of Melos)

Built from: Wikipedia — Cherry picking · Wikipedia — List of fallacies · Internet Encyclopedia of Philosophy — Fallacies · Logically Fallacious — Cherry Picking

Circular Definition

Explaining a word by using the word itself (or a twin of it), so you end up exactly where you started.

In a group chat about diets, someone asks, 'wait, what actually counts as a "clean" food?' A friend replies, 'easy — clean food is just food that isn't dirty. And dirty food is whatever isn't clean.' A few people react with a thumbs up like that settled it. But nobody in the chat is any closer to knowing whether the granola bar in their hand counts as clean or not.

The tell: The explanation sounds finished and confident, and your brain relaxes because a definition-shaped sentence got said. But the moment you try to actually use it — to decide one real case, like the granola bar — you notice you're holding two words that only point at each other, and you still have no way to tell them apart.

The move: Term T is defined as 'D', where D itself is T, a bare synonym of T, or resolves back to T within one or two further definitional steps (T =def D1; D1 =def T, or T =def D1; D1 =def D2; D2 =def T). The definiens (the defining phrase) presupposes prior grasp of the definiendum (the term being defined) instead of explaining it in independent terms.

Why it fails: A definition is supposed to let someone who does not already grasp a term come to understand it by way of terms they do already grasp. A circular definition substitutes the very term (or its synonym, or a short closed loop of mutually defining terms) back into its own explanation. Substituting the definiens for the definiendum anywhere it appears yields a sentence that still contains the undefined term, so no new information has been transmitted: the loop can be traversed endlessly without ever reaching ground. It becomes a live fallacy, rather than a merely poor dictionary entry, when the empty definition is then deployed to do argumentative work — treated as though it independently settles a contested case, when in fact it only restates the question.

• If you substitute the definition back in wherever the term appears, does the resulting sentence still contain the term itself or a bare synonym of it?
• Could someone who has never encountered this term learn to correctly apply it to new cases using only this definition, or must they already know what the term means?
• Trace the definition through each linked term one step at a time — does the chain loop back to the starting term within a step or two (its 'depth'), rather than bottoming out in independently understood terms?
• Is this definition just sitting in a dictionary, or is it being invoked right now to settle a contested case — is its emptiness doing real work in an argument?

Legitimate when: Circularity is not automatically vicious. Recursive definitions in mathematics and logic (e.g., defining a natural number as '0, or the successor of a natural number') are circular in surface form but legitimate because they bottom out in an independent base case and can be shown equivalent to a non-circular 'normal form.' Some theorists (revision theory) further argue that certain essentially circular definitions can still be meaningful, giving substantial guidance on a term's use across most cases even without a fixed, non-circular paraphrase. And any finite dictionary is full of small, unavoidable definitional circles among commonly understood terms (e.g., defining a 'nimbus cloud' as a cloud that produces precipitation) that cause no confusion and do no argumentative harm, because the circle closes among terms the audience already understands independently of the entry itself.

Ancient: Meno offers a third attempt at defining virtue: virtue is the power of attaining good things justly, where 'justly' is explained as acting virtuously. Socrates objects that this defines virtue by way of a part of virtue itself, so the search for what virtue is has made no progress and must start over. — Plato, Meno

Built from: Wikipedia · Stanford Encyclopedia of Philosophy — Definitions · Logically Fallacious — Circular Definition

Circular Reasoning

also known as: circulus in probando, vicious circle

Someone 'proves' a claim by using that very claim as the reason for it, just dressed in different words.

At the office, a coworker is pushing everyone to switch to a new project-management app. 'Trust the five-star rating,' she says. 'Why should I trust the rating?' you ask. 'Because it's rating a top-tier app — obviously a tool this good gets rated well.' 'Okay, but why do you believe it's a top-tier app in the first place?' 'Because look at that five-star rating!' You've gone all the way around the table and landed back where you started.

The tell: You ask 'why' once and get an answer, but when you ask 'why' about that answer, it loops back and hands you the original claim again, just wearing a different outfit.

The move: Claim P is supported by claim Q; claim Q is in turn supported by claim P (or, in the two-step case, P is supported directly by a restatement of P itself). The chain of support closes on itself instead of terminating in an independent ground.

Why it fails: The move is not formally invalid — P entails P, so nothing about the inference pattern itself is broken. The defect is probative: the premises are exactly as much in need of support as the conclusion, so the argument gives no independent reason to accept anything. Someone who did not already accept the conclusion has been given no new ground to do so; someone who did already accept it needed no argument at all. As Aristotle put it, if P were proved through Q and Q through P, things would in effect be 'proved' through themselves and nothing would end up better known than anything else.

• Trace each claim's support back through the chain: does it ever touch ground outside the loop, or does it eventually point back to the very claim being defended?
• Delete the conclusion from the argument entirely — does any remaining premise still have a source of justification, or do they all go dark?
• Can the 'reason' be restated without using the disputed term or a synonym for it anywhere in the restatement (e.g., swap 'trustworthy' for 'beyond reproach' and see if anything was actually added)?
• Ask the arguer to justify the first premise to someone who does not already grant the conclusion — do they end up quoting the conclusion back, just reworded?

Legitimate when: Mutual support among a network of claims is not automatically vicious: coherentist justification (where beliefs support each other as a web) can carry real evidential weight when the network is also anchored by claims with independent standing outside the loop, or when the loop is large, diverse, and constrained by outside checks (e.g. successful prediction, cross-checked observation). It becomes the fallacy specifically when the 'independent' support turns out, on inspection, to be nothing but the conclusion itself in different words, with no outside anchor anywhere in the chain.

Ancient: Aristotle argues that demonstration cannot proceed in a circle: if P were proved through Q and Q through P, things would in effect be proved through themselves, and nothing would be better known than anything else — genuine demonstration must instead descend from premises that are prior to and better known than the conclusion. — Aristotle, Posterior Analytics, Book I

Built from: Wikipedia — Circular reasoning · Wikipedia — Begging the question · Internet Encyclopedia of Philosophy — Fallacies · Stanford Encyclopedia of Philosophy — Fallacies

Commutation of Conditionals

Knowing that one thing leads to another doesn't mean seeing the second thing proves the first one happened.

In the group chat, Mia types: "If you catch the flu, you're wiped out for days." A few days later Jordan is dragging after finals week and says, "Guess I caught the flu then." But he'd just pulled three all-nighters in a row — being wiped out doesn't only happen because of the flu, so his flu claim doesn't actually follow from Mia's original comment.

The tell: It feels like the person quietly flipped the rule around — turning "this can cause that" into "seeing that means this happened" — without ever checking whether anything else could have caused the same result.

The move: P → Q; ∴ Q → P (the antecedent and consequent of a conditional are swapped and the result is treated as following from the original claim)

Why it fails: A conditional P → Q asserts only that P is sufficient for Q; it says nothing about what else might be sufficient for Q. Swapping the two claims, in effect, that P is also necessary for Q — a separate and stronger claim that was never established. Any row of the truth table where Q is true and P is false satisfies the original conditional while refuting the converse, so the move is invalid in every standard system of logic. The only relation that licenses the swap is material equivalence (P ↔ Q), which an ordinary 'if P then Q' does not assert.

• Can you name a different way Q could become true that does not go through P? If yes, the converse fails.
• Is the arguer quietly treating 'if P then Q' as if it meant 'if and only if P then Q' — and if so, where was that stronger equivalence actually argued for?
• Assume Q is true and P is false: does that violate anything the original conditional said? If not, the reversed claim has no support.
• Has the 'only if' direction (Q → P) been established on its own, independently of the 'if' direction (P → Q) that was actually argued?

Legitimate when: Only when the original relationship is a genuine biconditional (P ↔ Q) — P is both sufficient and necessary for Q — and that two-way equivalence has been independently established (by definition, stipulation, or by ruling out every other route to Q), rather than assumed from the one-way conditional alone.

Ancient: Aristotle observed that people often treat a one-way consequence as if it ran both ways: because the ground is wet after it rains, they assume that whenever the ground is wet, it must have rained. But the relation does not reverse — the ground could be wet from a burst pipe, dew, or someone washing a car, so wetness alone does not prove rain. — Aristotle, On Sophistical Refutations — discussed in the Stanford Encyclopedia of Philosophy's entry on fallacies as the 'fallacy of the consequent'

Built from: Wikipedia — Affirming the Consequent · LogicallyFallacious.com — Commutation of Conditionals · Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — List of fallacies

Complex Question Fallacy

also known as: plurium interrogationum, many questions fallacy, loaded question

It's a question that sneaks in an assumption about you that nobody ever agreed was true — so any way you answer, it sounds like you admitted it.

In the roommates' group chat, someone types: "Can we all agree it's time you stopped eating my leftovers without asking?" Nobody has ever actually shown that this person has been doing that. But now, answering "yes" sounds like admitting they've been sneaking food, and answering "no" sounds like they're refusing to stop doing it. There's no reply in the chat that doesn't make them look guilty.

The tell: That trapped, slightly unfair feeling — like you got handed a question where every single answer makes you look bad, and you can't figure out how that happened.

The move: A question Q is posed whose felicitous answering presupposes some proposition P. P has not been independently established, proven, or granted by the respondent. Every direct answer available under the question's format (typically 'yes' or 'no') is then treated as if it conceded P, because the question's wording gives the respondent no way to answer without appearing to accept P.

Why it fails: A question is not itself an argument, so it cannot legitimately transfer commitment to P the way a supported premise can. The defect is that the question format forecloses the very move a fair exchange requires — contesting P on its own before answering anything that depends on it. Wikipedia's account of the complex-question fallacy notes this is context-dependent: presupposing something is not automatically fallacious (many ordinary questions presuppose uncontroversial facts), so the failure specifically occurs 'when someone asks a question that presupposes something that has not been proven or accepted by all the people involved.' Stanford Encyclopedia of Philosophy traces the structural root to Aristotle, who classed this among the fallacies not dependent on language: 'Aristotle's fallacy of many questions occurs when two questions are asked as if they are one proposition' — combining a claim-question and a follow-up question into a single grammatical unit so that answering the follow-up looks like it settles the claim too.

• What must already be true — the hidden presupposition — for this question to even make sense as it's worded?
• Has that presupposition actually been stated as its own claim and agreed to, or is it just buried inside the question's grammar?
• Do all the direct answers this question allows (yes and no alike) still concede the presupposition either way?
• Can the question be split into a first question that asks about the presupposition on its own, and only then a second question that depends on the answer to the first?

Legitimate when: The move is legitimate when the presupposition is already established, uncontroversial, or explicitly stipulated by the person answering — for instance, 'Who is the monarch of the United Kingdom?' presupposes there is a monarch, which is common ground, whereas 'Who is the King of France?' presupposes a false claim (per Wikipedia's contrast pair). It is likewise fair in a courtroom once the underlying fact has been admitted: Wikipedia notes the same question can be 'loaded in one context, but not in the other,' depending on whether the defendant has already admitted the conduct in question.

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Complex question · Wikipedia — Loaded question · Internet Encyclopedia of Philosophy — Fallacies

Conflicting Conditions

Throwing out two or more excuses or promises that can't both be true, hoping the sheer number makes it sound more convincing instead of less.

A teenager explaining the new dent in the family car: "I didn't drive it last night. Also, that dent was already there before I even had the keys. Also, somebody must have hit it while it was parked and there was nothing I could've done." Each sentence sounds like a reasonable thing to say on its own. But notice: if the car was never driven, nobody driving it could have caused a dent last night. If the dent was already there, the parked-car story is beside the point. Offer all three and you haven't tripled your defense — you've quietly admitted you don't have one that holds up.

The tell: You get handed several different explanations in a row for the exact same thing, and each one is being offered as if it's the full answer on its own — but the moment you'd need to compare them side by side, one of them has to give way. It feels like being reassured by a bouquet of alibis instead of one steady one.

The move: A position P is defended by asserting (or silently presupposing) that conditions C1, C2, ... Cn must all hold. But C1 through Cn are jointly unsatisfiable — no assignment of truth values, and no single state of affairs, can make them all true together. The arguer nonetheless treats P as secured by the whole set, rather than noticing that at most one subset of compatible conditions could ever obtain.

Why it fails: Because the conjunction (C1 & C2 & ... & Cn) is a contradiction, there is no possible world in which the full defense is true. Any conclusion said to rest on 'all of the above' is therefore resting on nothing: a contradiction does not describe a weaker or more robust position than a single claim, it describes no position at all. Freud's classic case (the borrowed-kettle excuses, in The Interpretation of Dreams and Jokes and Their Relation to the Unconscious) shows the seductive move: each excuse looks individually plausible, and stacking them feels like it multiplies the odds of acquittal, but since 'I never borrowed it,' 'it was already cracked,' and 'I returned it undamaged' cannot all be true of the same kettle, the stack proves nothing beyond that at most one of them might be.

• List every condition the position actually needs to be true. Can all of them hold at once, even in principle — or does asserting one quietly rule out another?
• Swap any one required condition for its opposite: does the rest of the defense collapse or become unnecessary? If so, the conditions were doing contradictory work all along.
• Does the apparent conflict dissolve under a charitable reading — different times, different scopes, different senses of an ambiguous term? If yes, this is not conflicting conditions; it is equivocation or a merely apparent contradiction.
• Is each incompatible condition being handed to a different audience or a different moment, so that no single listener ever sees them side by side (the 'kettle logic' multiple-defense pattern)?

Legitimate when: The technique of deliberately deriving a contradiction is legitimate as reductio ad absurdum: there, the arguer explicitly adopts an opponent's premises in order to show they entail a contradiction, and then rejects them for that reason — the contradiction is the point, not a hidden defect. It is also not this fallacy when the 'conflict' is only apparent because the conditions apply at different times, in different respects, or under different senses of a term (e.g., 'lower rates now, raise them later' is not self-contradictory, since 'now' and 'later' pick out different states of affairs).

Built from: Wikipedia — Kettle logic · Internet Encyclopedia of Philosophy — Fallacies (entry: Inconsistency, under Fallacies of Illegitimate Presumption) · Stanford Encyclopedia of Philosophy — Fallacies

Confusing an Explanation with an Excuse

Saying why someone did something wrong gets twisted into 'so you think it's fine that they did it.'

In the team group chat, Maya writes: 'Jordan missed the deadline because his ticket got stuck behind an infrastructure outage nobody on the team could even see.' Priya fires back: 'Wow, so you're saying it's totally fine he missed it? Must be nice getting a free pass.' Maya never said it was fine. She said why it happened. Priya answered a question nobody asked.

The tell: You suddenly feel accused of a stance you never took, and you catch yourself saying 'I didn't say it was okay - I said why it happened,' even though you're pretty sure that's exactly what you did say the first time.

The move: S asserts a descriptive/causal claim: 'A occurred because of C' (an explanation of why A happened). B infers, from that claim alone, the normative claim 'S holds that C makes A permissible / not blameworthy' (an excuse), and then attacks S for holding that unstated normative position. The inference from 'S named a cause of A' to 'S is excusing A' is not licensed by anything S actually said; the descriptive claim and the normative verdict are logically independent, so B has swapped S's real (descriptive) claim for a different (normative) one before attacking it.

Why it fails: An explanation answers 'why did this happen' (a causal/descriptive question); an excuse answers 'should this reduce or remove blame' (a normative question). The truth of the causal claim is entirely compatible with condemning the act in the strongest terms - citing a cause neither entails nor even suggests approval. Because the critic responds to the normative claim 'A was permissible' rather than to the descriptive claim actually made, the reply is irrelevant to what was said; once the substitution happens, the critic is typically attacking a position the explainer never took, which collapses the move into a straw man.

• Is the original claim descriptive (why A happened) or normative (whether A was permissible or blameless)? Quote the actual words used, don't paraphrase them into a verdict.
• Would the causal explanation remain exactly as true if the speaker also condemned the act in the strongest possible terms? If yes, the explanation cannot be doing excusing work.
• Has the critic pointed to any separate statement where the explainer said the act was justified, forgivable, or not the person's fault - or is the 'excusing' charge inferred solely from the fact that a cause was named at all?
• If the cited cause were removed from the story, would the act become any less wrong - or merely more mysterious? A cause that only makes the act less surprising, not less bad, is explanatory, not exculpatory.

Legitimate when: The move is not fallacious when the explainer's own words, or an established legal/moral framework, actually use the cause as grounds for reduced responsibility - e.g., doctrines of duress, insanity, provocation, or diminished capacity, where a causal story is offered precisely to argue for a lighter verdict. In those settings 'he explained it' really does mean 'he is asking for leniency,' because the explanation and the excuse have been deliberately fused by the speaker; treating that fused claim as an excuse is simply reading it correctly, not committing the fallacy.

Ancient: Discussing which acts count as involuntary and so deserve pardon, Aristotle considers acts done out of anger or appetite and rules them voluntary rather than involuntary, since they arise from within the agent rather than from outside force or blameless ignorance. Naming anger or desire as the cause of an act, on his account, does not turn it into something excusable the way genuine compulsion does - the cause explains the act without discharging the responsibility for it. This is the philosophical bedrock underneath the fallacy: Aristotle already treats 'what caused it' and 'does it merit pardon' as separate questions with separate answers. — Aristotle, Nicomachean Ethics, Book III, Chapter 1

Built from: Internet Encyclopedia of Philosophy - Fallacies (entry: Confusing an Explanation with an Excuse) · Wikipedia - Bulverism · Wikipedia - Genetic fallacy · Stanford Encyclopedia of Philosophy - Fallacies · Aristotle, Nicomachean Ethics, Book III (MIT Classics Archive translation)

Confusing Currently Unexplained with Unexplainable

Just because nobody has figured something out yet doesn't mean nobody ever will — but this trick treats 'not explained yet' as if it were 'can never be explained.'

Your uncle's smart thermostat keeps dropping offline at 3 a.m. He calls tech support, who can't figure out why after one call. Uncle announces at dinner: "They can't explain it. Nobody can explain it. This isn't a wifi glitch — something else is going on in that house," and now he's telling everyone he thinks the router is haunted, or the neighbor is jamming the signal on purpose. One unanswered support ticket became proof that ordinary causes are off the table.

The tell: Someone treats 'nobody has explained this yet' as if it were the same thing as 'this can never be explained' — and then quietly slides a much stranger, more dramatic story in to fill the space, as though that story just won by forfeit.

The move: Phenomenon X currently has no accepted explanation (no one has yet produced one, or the available explanations are contested). Therefore X cannot be explained, in principle, by ordinary causes. Therefore some preferred exotic, supernatural, or otherwise extraordinary explanation Y is warranted by default.

Why it fails: The premise reports a fact about the present state of an inquiry (an epistemic gap: nobody has yet supplied an adequate explanation). The conclusion asserts a fact about the world's structure (a modal claim: no adequate explanation exists to be found). Nothing in the premise licenses that jump — a gap in current knowledge is evidence about the state of current knowledge, not about the limits of what is knowable. The move is a variant of the argument from ignorance with an extra modal escalation: first 'not yet explained' is smuggled into 'unexplainable,' and then that manufactured impossibility is used to install a preferred alternative by default, without that alternative ever having to earn its keep on positive evidence. The history of inquiry supplies the standing counterexample: lightning, disease, the orbits of planets, and countless other once-mysterious phenomena were all 'currently unexplained' before they were explained, so 'currently unexplained' has repeatedly turned out not to entail 'unexplainable.'

• Does the argument move from 'no one has explained X yet' to 'X cannot be explained'? What licenses that jump from a gap in knowledge to a claim about what is knowable?
• How hard has anyone actually looked — is the absence of explanation the product of serious sustained investigation, or simply of nobody having tried yet?
• Is 'unexplainable' being used to hand victory to a preferred alternative explanation by default, rather than that alternative earning support through its own positive evidence?
• If a mundane explanation for X were found next year, would that surprise the arguer — or does their own confidence suggest they already suspect X is merely unexplained-so-far rather than truly beyond explanation?

Legitimate when: It is legitimate to treat something as very likely permanently unexplained only after a genuinely exhaustive, competent search has failed and there is independent, positive reason (not just failure-to-find) to think no explanation is available in the relevant framework — and even then the honest conclusion is a hedged 'we have found no explanation and have principled reason to expect none,' not a flat, triumphant 'unexplainable' pressed into service to justify a specific preferred alternative.

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Argument from ignorance · Wikipedia — List of fallacies · Wikipedia — Argument from incredulity · Wikipedia — God of the gaps · Internet Encyclopedia of Philosophy — Fallacies

Conjunction Fallacy

also known as: Linda problem

A story with more specific details can feel more likely than a plain, general one — even though extra details can only ever make something less likely, never more.

In the group chat, someone asks why Priya bailed on brunch. One friend says, 'she probably overslept.' Another jumps in with, 'no, she overslept AND her phone died overnight AND she couldn't find her charger to set a backup alarm.' Everyone reacts to the second version with 'oh yeah, that's definitely what happened' — the extra detail makes it feel more true, even though it now requires three separate things to have gone wrong instead of one, which makes it the less likely story of the two, not the more likely one.

The tell: A version of the story that's easier to picture, or that fits the person's personality better, suddenly feels more 'likely' than the plain version — even though the plain version already covers everything the detailed version covers, plus more besides.

The move: Let A and B be two claims. The arguer is asked to rank the likelihood of A against the likelihood of the compound claim that A and B both hold, and ranks the compound higher: P(A ∧ B) > P(A). Every situation in which A and B both hold is already a situation in which A holds, so the compound picks out a subset of the cases the bare claim picks out. Adding the conjunct B can only shrink the qualifying set, never enlarge it; but B is a detail that makes the story cohere, and the arguer substitutes how well the compound matches a mental picture for the measure of the set it names.

Why it fails: Every situation in which A and B are both true is also a situation in which A is true, so the set of A-and-B outcomes is always a subset of (or equal to) the set of A outcomes. By the axioms of probability this makes P(A ∧ B) ≤ P(A) a hard ceiling, not a rule of thumb. When someone rates the conjunction higher, they have stopped tracking probability and started tracking how well the detailed story matches a mental stereotype (the 'representativeness heuristic' Tversky and Kahneman used to explain the Linda experiments) or how coherent and vivid the narrative feels.

• Can option B be restated as 'A, and also C,' where C is a separate added condition? If so, B is a subset of A and cannot legitimately outrank it.
• Would drawing the two options as circles show one fully inside the other? If the 'more probable' one is the smaller, inner circle, the judgment is already contradicted.
• Is the extra detail doing narrative work (making the story feel more representative or true-to-character) rather than evidential work (actually raising the odds)?
• Could the audience be silently reading the broad option as excluding the detail — i.e., hearing 'bank teller' as shorthand for 'bank teller and not a feminist activist' — so the comparison isn't really A vs. A-and-B at all? This conversational reading would make the error one of ambiguous wording rather than of probability.

Legitimate when: Never valid under standard probability theory once A and B are interpreted consistently — a conjunction cannot legitimately be judged more probable than either of its own conjuncts. The one live scholarly caveat: some experimental results (raised by critics such as Gerd Gigerenzer) may reflect participants reading the plain option as tacitly excluding the added detail, or reading 'probable' loosely, rather than a genuine probability error; framing questions in frequency terms ('how many out of 100') reduces but does not eliminate the effect, which is why researchers still treat the core pattern as a real error, not just an artifact of wording.

Built from: Wikipedia — Conjunction fallacy · Wikipedia — List of fallacies · Internet Encyclopedia of Philosophy — Fallacies

Conspiracy Theory

A story rigged so nothing could ever prove it wrong — evidence for it counts, and missing evidence just proves it's being hidden.

My coworker Dana is convinced our whole team is secretly getting laid off next month. I show her the email HR sent confirming everyone's job is safe. "Of course they'd say that," she says, "they don't want a panic before they pull the trigger." I point out there's no hiring freeze, no budget cuts, nothing unusual going on anywhere. "That's exactly how they'd hide it," she says. "If it looked obvious, it wouldn't be secret." Whatever I bring her — reassurance or silence, good news or nothing at all — somehow turns into more proof she was right all along.

The tell: You realize there is nothing you could possibly show her that would change her mind. Good news, bad news, and no news all get absorbed into the same story, which means the story was never really about the evidence in the first place.

The move: Claim: hidden group G secretly caused/controls event E. For any evidence X encountered: if X appears to support the G-hypothesis, X is cited as confirmation; if X appears to contradict it, X is reinterpreted as itself part of G's concealment effort; if X is simply absent (no documents, no witnesses, no leaks), that absence is read as proof of how thoroughly G covered its tracks. Therefore: G-hypothesis is true, and (unstated but required) no possible observation could ever count against it.

Why it fails: The argument converts every logically possible outcome — confirming evidence, disconfirming evidence, and no evidence at all — into support for the same conclusion. A claim that no observation could ever disconfirm has been insulated from the very evidential process that was supposed to establish it; per Popper's falsifiability criterion (Wikipedia, Falsifiability), a theory that 'explains everything' by re-interpreting apparent refutations to fit — Popper's 'conventionalist twist,' seen when 'the followers of Marx re-interpreted both the theory and the evidence in order to make them agree' — has traded evidential support for a structure that merely presupposes its own truth and then absorbs everything else as ornamentation. This is a presumption fallacy: the hidden-group hypothesis is assumed at the outset and used to reinterpret the evidence that was meant to test it, rather than being established by evidence that could have come out otherwise.

• What observation, if any, would the proponent accept as evidence against the conspiracy claim?
• Is absence of supporting evidence being cited as proof of concealment rather than as evidence of nothing to conceal?
• Does every apparent disconfirmation get folded back in as further proof the conspiracy is real (e.g., 'that's exactly what they'd want you to think')?
• Does the claimed conspiracy's required scale — the number of coordinated, silent participants sustained over the relevant time — survive a basic plausibility check?

Legitimate when: Positing a hidden coordinated actor is not itself the defect — real conspiracies (price-fixing rings, cover-ups, intelligence operations) exist and have been established by ordinary evidence: documents, leaked communications, testimony, financial records. The reasoning is not fallacious when the claim is falsifiable in practice — the proponent can specify what would count against it — and rests on evidence gathered independently of assuming the conspiracy's truth, evaluated the same way any other historical or factual claim would be. As the Internet Encyclopedia of Philosophy notes, some initial resistance to falsification is expected of any genuinely secret plot and is not on its own damning; what marks the fallacious version is that disconfirmation and its absence are treated as equally, automatically confirming, which is the same move a 'degenerating research program' makes when 'auxiliary hypotheses and initial conditions are successively modified in light of new evidence, to protect the original theory from apparent disconfirmation.'

Built from: Internet Encyclopedia of Philosophy — Conspiracy Theories · Wikipedia — Conspiracy theory · Wikipedia — Falsifiability

Contextomy

Someone clips a few of your exact words out of a longer sentence so it sounds like you said the opposite of what you meant.

In the group chat, Priya types: "honestly the new schedule stresses me out, but I'll manage, it's fine." Later, Dev screenshots just "the new schedule stresses me out" and posts it to the team channel to prove Priya hates the new schedule and everyone should push back on it.

The tell: The words are technically hers, so it feels unfair to call it a lie — but something about how neatly it fits the screenshot-maker's point feels staged, like a movie trailer that makes a bad film look like a masterpiece using three good words strung together.

The move: Speaker S utters passage P within context C, where C fixes P's meaning as M. An excerpt E is cut from P such that E reproduces S's exact words but, once severed from C, is presented as asserting M', where M' ≠ M (typically M' is more extreme, more damning, or more supportive of the excerpter's aims than M). The excerpt is then attributed to S as though S asserted M'.

Why it fails: The defect is semantic, not evidentiary: the quoted string is accurate, but the proposition it is used to assert is not the one the original utterance expressed. Meaning in natural language is context-dependent — qualifying clauses, negations, 'but' statements, and surrounding sentences fix reference and scope. Removing that surrounding material lets the same string carry a different, unendorsed proposition. This is why Wikipedia's List of fallacies files contextomy/quote mining under the equivocation family of informal fallacies rather than under relevance defects: the excerpter trades on the ambiguity created by decontextualization, then treats the resulting (distorted) meaning as though it were the settled one.

• Read the full passage: does the surrounding context reverse, qualify, or reattribute the quoted clause (e.g., the speaker was voicing an opponent's view, or a caveat immediately follows)?
• Does the excerpt cut mid-sentence, or omit a conditional, negation, or 'but/however' clause that governs it?
• Would the argument survive if the complete quotation were substituted for the excerpt?
• Is the quoted speaker being made to say something more extreme, more damning, or more convenient for the excerpter's case than the full passage supports?

Legitimate when: Excerpting is legitimate — indeed unavoidable — whenever the trimmed material does not change what the passage asserts: e.g., quoting a self-contained sentence whose meaning does not depend on adjacent qualifiers, or quoting a source's actual bottom-line conclusion while leaving out redundant supporting detail. The test is substitution: if reinserting the full passage would not alter the point being attributed to the speaker, the excerpt is a fair quotation rather than contextomy.

Built from: Wikipedia — Quoting out of context · Wikipedia — List of fallacies · Stanford Encyclopedia of Philosophy — Fallacies

Deceptive Sharing

Taking a real photo, video, or quote and slapping a new caption on it that makes it 'prove' something it never actually showed.

In a parent group chat, someone posts a photo of moldy strawberries in a clamshell container: "These are from FreshMart on 5th St, bought TODAY — be careful!" The photo itself is completely real, nobody edited it, those strawberries really did go moldy. But the picture was actually taken at a farm stand three states away, four years ago, originally shared as a tip about home composting. Nobody in the chat checks. By dinnertime half the neighborhood has sworn off FreshMart strawberries.

The tell: That nagging feeling of 'okay, but how do we actually know that's when and where this happened?' The picture checks out — nobody's faked it — but nobody has actually confirmed what it's a picture OF.

The move: Content C (a photo, video, document, or quotation) is genuine and independently verifiable. C is shared together with a framing F — a caption, headline, timestamp, or attribution — that asserts C depicts, proves, or is connected to proposition P (a specific time, place, event, or cause). The audience reasons: 'C is real' + 'F says C establishes P' → accept P. The inference is invalid because C's authenticity says nothing about whether F's claim about C is true; the link between C and P is an unstated, unsupported premise smuggled in by the framing rather than by the content itself.

Why it fails: The move launders a false or unsupported claim through the credibility of something true. Verifying that the photo is real, the video unedited, or the quotation accurately transcribed does nothing to verify the separate claim about what that content shows, when, where, or why — yet the audience's scrutiny typically stops at 'is this real?' and never reaches 'does this real thing actually show what I'm being told it shows?' The presumption that the framing accurately reports the content's context is asserted, not argued for.

• Is the underlying content itself authentic and unaltered — and separately, is its claimed time, place, and source verifiable, or only asserted?
• Does the caption, headline, or framing assert a connection (a date, a location, a cause, an identity) that the content itself does not establish on its own?
• Who supplied the caption, and is that the same person or institution that actually created or originally published the content?
• If you trace the content back to its original context (reverse image search, full transcript, original publication date), does the claimed connection survive, or collapse?

Legitimate when: Sharing genuine content is not fallacious when the attached framing accurately reports the content's real provenance and the connection it asserts is the one the original source itself documented or intended — that is simply reporting, not deceptive sharing.

Built from: Wikipedia — Fake news (false context, Claire Wardle / First Draft typology) · Wikipedia — Cherry picking · Wikipedia — Quoting out of context · Wikipedia — Misinformation (malinformation distinction) · Wikipedia — List of fallacies · Stanford Encyclopedia of Philosophy — Fallacies

Definist Fallacy

Someone quietly swaps what a word means so that just by using their new meaning, they've already won the argument -- no evidence needed.

In the group chat about splitting a dinner bill, Mia says: 'Look, being "fair" just means everyone pays the exact same amount, no matter what they ordered.' Then: 'Since fair means equal shares, and equal shares is what I'm asking for, you basically have to agree I'm right.' Nobody ever agreed that's what 'fair' means -- she just announced it, and then used her own announcement as her whole case.

The tell: You notice the argument has quietly stopped being about the actual situation and turned into an argument about a word -- and somehow agreeing to their meaning of the word means agreeing to their entire point.

The move: Term T is contested. Arguer stipulates a new definition: 'T means D' (D chosen so it is favorable or unfavorable to a position the arguer already holds, while keeping T's ordinary praise/blame charge). Arguer then argues: X is D; therefore X is T (and inherits whatever approval or condemnation attaches to T) — or, in the negative form: X is not D (as redefined); therefore X is not T.

Why it fails: The move looks like a harmless clarification of terms but is actually a substantive, contested claim smuggled in under the cover of a definition. A genuine definition should be neutral ground both sides of the dispute would already grant (lexical, technical, or explicitly stipulated by agreement); here the definition is rigged so that the very question at issue is settled by fiat rather than by evidence or argument. Because the redefined term keeps its original emotional charge even as its descriptive content is swapped out, the listener who grants 'the definition' unknowingly grants the conclusion too — the persuasive work is done entirely by word choice, not by any reason offered.

• Would the other side of this dispute accept the definition just offered, or only someone who already holds the arguer's conclusion?
• If you substitute a neutral, dictionary, or standard technical definition of the key term, does the conclusion still follow?
• Is the definition doing evidential work — settling a factual or contested question — that only observation or argument could legitimately do?
• Does the term keep its original positive or negative charge even though its actual content has been quietly changed?

Legitimate when: When the definition on offer is the term's established lexical, technical, or openly stipulated-and-agreed meaning rather than a redefinition engineered to preload the conclusion. Explicit, transparent redefinition for a limited purpose — as when a statute, contract, or scientific paper states 'for purposes of this document, X means Y' and both sides can inspect and contest that stipulation on its own terms — is a legitimate technique, not the fallacy; the fallacy requires that the tendentious definition be passed off as the term's ordinary or uncontroversial meaning.

Ancient: Thrasymachus declares that 'justice' simply means whatever serves the interest of the stronger party. Once that stipulation is granted, his conclusion -- that the powerful are entitled to disregard ordinary standards of justice -- seems to follow automatically, even though the definition itself is exactly what is in dispute. Socrates spends much of Book I contesting the stipulated meaning rather than accepting it and arguing about the conclusion. — Plato, Republic, Book I (Thrasymachus on justice)

Built from: Wikipedia - Persuasive definition · Wikipedia - Definist fallacy (logic) · LogicallyFallacious.com - Definist Fallacy · Stanford Encyclopedia of Philosophy - Plato's Ethics and Politics in The Republic

Denying a Conjunct

Just because two things can't both be true at once doesn't mean the one you didn't rule out has to be true — it's entirely possible neither one is.

A parent texts about a jacket: "The tag says this thing can't survive being both machine-washed and put in the dryer — one or the other wrecks it. You didn't put it in the dryer. So you must have machine-washed it." Except the kid just left it balled up, damp, on the closet floor for three days. Nobody washed it and nobody dried it. Ruling out the dryer told us nothing about the washer.

The tell: That nagging "wait, couldn't it just be neither?" feeling — like you're being herded toward one of two doors when a third option (neither) was standing open the whole time.

The move: Two claims P and Q are governed by a premise ruling out their joint truth: ¬(P ∧ Q) — not both. The arguer denies one of the two conjuncts and concludes the other: ¬(P ∧ Q); ¬P; ∴ Q, and symmetrically ¬(P ∧ Q); ¬Q; ∴ P. The substitution happens at the connective. A NAND excludes exactly one of the four assignments of truth values to P and Q — the one where both are true — leaving three standing. An exclusive disjunction would exclude two, both-true and both-false. The arguer reads the first as though it were the second, and the both-false row, which the stated premise leaves entirely untouched, is precisely the row on which the conclusion fails.

Why it fails: The premise 'not both P and Q' rules out only the single case where P and Q are both true. It says nothing about the case where P and Q are both false. So learning that P is false does not pin down Q at all — Q can be true or Q can be false, and 'not both P and Q' is satisfied either way. The argument silently smuggles in an assumption it never earned: that at least one of P or Q has to be true (an exclusive-or), when all it was actually given was a NAND (not-both).

• Can I describe a situation where both P and Q are false, and nothing in the stated premises is violated by that?
• Is 'not both P and Q' quietly being treated as if it meant 'exactly one of P and Q'?
• Has 'at least one of these must be true' actually been established somewhere, or is it just being assumed by the shape of the sentence?
• If the other conjunct had been denied instead, would the argument have 'proven' the opposite conclusion with equal confidence?

Legitimate when: The move becomes valid only when an additional, independently established premise guarantees that at least one of P or Q must hold (i.e., P ∨ Q is separately known, not merely assumed). Combined with 'not both,' that upgrades the situation to a genuine exclusive alternative, and denying one conjunct then validly yields the other by disjunctive syllogism. Note this is a different, and already valid, pattern: 'not both P and Q; P; therefore not Q' (affirming one conjunct to deny the other) is sound modus-tollens style reasoning and should not be confused with the fallacy, which runs the invalid direction — denying one to affirm the other.

Built from: Logically Fallacious (Bo Bennett) — Denying a Conjunct · The Fallacy Files — Denying a Conjunct · Wikipedia — Affirming a Disjunct

Denying the Antecedent

Ruling out one reason something could happen doesn't rule out the thing happening.

In the office group chat, Deja types: "If the printer's jammed, the handouts won't print." Marcus checks the tray — no paper stuck anywhere. "It's not jammed," he says. "Great," Deja replies, "so it'll print fine." Except the toner cartridge is bone dry, which stops the print job just as dead as a jam would have — and the room is still handout-less when the meeting starts.

The tell: That itch you get when clearing away one explanation gets treated like the whole problem is solved — the same feeling as when one suspect gets an alibi and everyone acts like the case is closed, instead of asking who else could have done it.

The move: If P then Q; not P; therefore not Q. (P → Q, ¬P, ∴ ¬Q)

Why it fails: The conditional "if P then Q" asserts only that P is sufficient to produce Q — it never claims P is the only thing that can produce Q. Denying P and concluding not-Q treats the one-way conditional as though it were reversible (as though P ↔ Q), which the original statement never established. Q can be true through some route that has nothing to do with P, so P's falsity places no logical constraint on Q's truth value; any case where P is false and Q is true refutes the inference outright.

• Is P being treated as the only way Q could come about, or just one way among several?
• Could Q still be true even though P is false — is there a rival route to the same outcome?
• List the other conditions that could produce Q on their own: does the argument actually rule them all out?
• Was the original claim really "P guarantees Q," or has it quietly been treated as "only P produces Q" (a hidden if-and-only-if that was never argued for)?

Legitimate when: The move is legitimate only when P is not merely sufficient but also necessary for Q — that is, when the converse (Q → P) independently holds, so the conditional is really a biconditional (P if and only if Q). This is common with stipulative definitions, formal rules, and narrowly bounded technical conditions where "if" is being used to mean "if and only if"; outside those cases, denying P tells you nothing about Q.

Built from: Wikipedia — Denying the antecedent · Wikipedia — List of fallacies · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies · Logically Fallacious — Denying the Antecedent

Denying the Correlative

Dodging a real yes-or-no question by inventing a fake third answer that isn't actually different from yes or no.

In the group chat, the host writes: "I need a headcount for the venue by tonight — are you coming Saturday or not, just yes or no." One friend replies: "I'm not really a 'no,' I'm more of a 'we'll see how the week goes.'" The host still has to write one number on the form for the venue, and "we'll see" isn't a box on it — it's just a soft way of not committing, dressed up to sound like a real third choice.

The tell: The answer never actually lands on yes or no — it just uses more words to avoid saying either one, even though the situation genuinely only has two boxes to check.

The move: A pair of terms P and not-P is genuinely exclusive and exhaustive relative to the question in force (a recorded yes/no vote, a guilty/not-guilty verdict, a did-you-do-it-or-not question). Asked to pick, the respondent answers with neither P nor not-P but with some R, implying a third position exists between or beside them — while R is in fact just P or not-P wearing a new label, or an answer to a different question entirely.

Why it fails: The rules of the exchange already fixed P and not-P as jointly exhaustive — a ballot only records yes or no, a verdict is guilty or not guilty — so 'neither' is not an available outcome. The apparent third option is manufactured, not discovered: strip its relabeling away and it collapses back into P or not-P, or it turns out to answer a question nobody asked. The correlative is denied in words but not actually escaped in fact, so the response fails to do the one thing a forced choice requires.

• Is the pair actually forced by the rules of the situation (a recorded vote, a verdict, a yes/no ledger entry) rather than just conversational habit?
• Strip the new label off the 'third option' — does it turn out to just be P or not-P restated in softer or vaguer words?
• Does the answer address the question that was actually asked, or does it quietly swap in an easier, adjacent question?
• Could a scorekeeper (a form, a ledger, a jury) record this answer as anything other than one of the two original options?

Legitimate when: When the offered dichotomy is not really exclusive and exhaustive — a genuine false dilemma — pointing to a third option is the correct, non-fallacious rebuttal, not this fallacy. It is likewise legitimate when the question's own presupposition is false (e.g., 'did you stop cheating?' asked of someone who never cheated): refusing to pick a horn is then the right move, because no true correlative pair was actually on the table.

Built from: Wikipedia — Denying the correlative · Logically Fallacious — Denying the Correlative · Wikipedia — False dilemma · Wikipedia — Correlative-based fallacies

Disjunction Fallacy

A specific, colorful guess can feel more likely than the plain, broad category it's actually part of — even though the big category can never come in below any one thing inside it.

Your dog is limping and the vet visit is tomorrow. Your partner says, 'I'd bet real money it's specifically that he twisted his ankle jumping off the couch — way more likely than just some vague "leg injury."' But 'leg injury' already covers twisted ankle, sprain, torn nail, thorn, sore paw pad, early arthritis — every one of those IS a leg injury. The narrow, vivid guess can't outrank the broad label that already contains it, no matter how sure your partner sounds.

The tell: The broad, catch-all option feels bland and made-up on the spot, so it seems less real than the one specific, colorful guess — even though the broad option quietly includes that guess as one of its own possibilities.

The move: A specific possibility A is judged more probable than a broader possibility (A or B) that logically contains it: Judged P(A) > P(A ∨ B), even though A entails A ∨ B.

Why it fails: By the axioms of probability, if A entails A-or-B — every situation in which A holds is automatically a situation in which 'A or B' holds too — then P(A) ≤ P(A ∨ B) is a hard floor, not a rule of thumb: the broader category can never trail behind one of the specific things it contains. When someone rates the narrow case as more likely anyway, they have stopped tracking probability and started tracking how well a vivid, detailed story matches a mental stereotype (the representativeness heuristic). The broad, catch-all category, having no story of its own, feels vague and therefore 'less likely,' even though mathematically it has to be at least as likely as any single possibility folded into it.

• Does the narrower option actually sit inside the broader one — is every case where the specific guess is true automatically a case where the broad option is true too?
• If you spell out the broad option's full reach explicitly ('any of these, including this particular one'), does the odds-ranking flip back the way it has to?
• Is a vivid, specific-sounding detail (a name, a motive, a mechanism) doing the persuading here, rather than any actual evidence bearing on likelihood?
• Is the comparison really 'narrow item vs. the broad category containing it,' or is it secretly 'this labeled option vs. that labeled option' where the two labels are meant to be mutually exclusive choices rather than nested categories?

Legitimate when: Never valid once the categories are interpreted consistently: if A truly is a subset of (A or B), the broader option cannot legitimately be judged less probable, full stop. The judgment stops being an error only if the apparent 'broader' option isn't actually a superset after all — e.g., a form or a person offers 'physics' and 'natural sciences' as two exclusive labels to pick between (not as narrow-inside-broad), in which case the comparison being made is a different one and the containment relation the fallacy depends on was never really there.

Built from: Wikipedia — Representativeness heuristic · Wikipedia — Conjunction fallacy · Wikipedia — List of fallacies · LogicallyFallacious.com — Affirming a Disjunct

Distinction Without a Difference

Getting out of trouble by renaming the exact same thing with a nicer-sounding word.

Your roommate leaves a pan crusted with last night's dinner in the sink for three days. You say, "Can you stop leaving dirty dishes in the sink?" They say, "I don't leave dirty dishes in the sink — I let them soak." It's the same pan, in the same sink, for the same three days. Nothing about what's actually happening changed; only the label did.

The tell: You feel like you just lost an argument you were clearly right about, and you can't immediately say why — because the other person answered a different-sounding sentence instead of the actual complaint.

The move: Speaker asserts that a rejected label L1 and an accepted label L2 name two different things, and treats that asserted difference as doing real argumentative work (exculpating a charge, blocking an inference, or licensing a different conclusion) — while nothing that would be true, permitted, owed, or predicted under L1 fails to hold under L2, and vice versa. The 'distinction' lives entirely in which word is chosen, not in any feature of the thing being described.

Why it fails: For a distinction to do argumentative work, affirming L2 must leave open the possibility of denying L1 — the two descriptions must diverge in truth conditions, so that some fact obtains under one but not the other. If L1 and L2 are truth-conditionally equivalent (everything that is true, owed, or predicted under one is true, owed, or predicted under the other), then denying L1 while affirming L2 denies and affirms the identical state of affairs under different names. The argument borrows the rhetorical force a genuine conceptual distinction would carry, while supplying only a change of vocabulary, so the conclusion the speaker wants (e.g. 'therefore I did nothing wrong') does not follow from any actual difference in the world.

• List what would be true, permitted, or predicted under label L1 that is NOT also true, permitted, or predicted under label L2 — is that list empty?
• If every occurrence of L1 were swapped for L2 throughout the situation (and vice versa), would any obligation, cost, penalty, or outcome actually change?
• Is the arguer trying to shed the emotional charge that attaches to L1 while quietly keeping everything L1 refers to intact?
• Running the check the other way: is the arguer dismissing a distinction as 'just semantics' when it would in fact trigger a different legal, moral, or practical consequence?

Legitimate when: The two labels really do diverge in truth conditions, obligations, or predicted consequences, even though they sound similar or are casually treated as synonyms in ordinary speech — for instance 'negligence' vs. 'recklessness,' or 'tax avoidance' vs. 'tax evasion,' each trigger different legal tests, burdens of proof, and penalties, so insisting on keeping them separate is substantive, not evasive wordplay.

Built from: Wikipedia — Distinction without a difference · Logically Fallacious — Distinction Without a Difference · Stanford Encyclopedia of Philosophy — Fallacies (equivocation/amphiboly classification, consulted for relation to ambiguity fallacies) · Internet Encyclopedia of Philosophy — Fallacies (equivocation entry, consulted for relation to ambiguity fallacies)

Double Standard

Using one rulebook to judge someone else and a completely different, easier rulebook to judge yourself (or your own side) for the exact same kind of thing.

In a family group chat, a dad texts his teenage son: "You're grounded for a week — I saw the video, you were doing 15 over the limit near the school." Twenty minutes later the same dad posts a photo of himself doing 80 on the highway with the caption "road trip time!" When his son points it out, the dad replies, "That's completely different, I'm an experienced driver." Nobody asked whether speed limits near a school care about driving experience — he just needed an excuse that let him off the hook.

The tell: You notice the same behavior gets called "reckless" in one person's hands and "no big deal" in another's, and when you ask why, the answer is vague, comes only after you push back, or basically boils down to "because it's me."

The move: Criterion C is applied to case A to reach a verdict (condemn, excuse, penalize, praise). The same criterion C, present in relevantly similar case B, is not applied — B is judged by a different, more lenient (or stricter) criterion C'. No relevant difference between A and B is stated or defended to justify using C for one and C' for the other.

Why it fails: The move fails not because either verdict is wrong in isolation, but because the reasoner presumes, without argument, that a difference exists which is relevant enough to license applying two different rules to what is otherwise the same kind of case. That presumed-but-undefended exempting distinction is the defect: the asymmetry is asserted, not shown. This makes it a fallacy of illegitimate presumption rather than a formal error — the argument's form is fine, but a needed premise (why A and B differ in a way that matters to C) is missing and, when demanded, usually turns out to be indefensible or ad hoc.

• State the criterion actually being used to judge the criticized case, in words general enough to also cover the speaker's own parallel case — does the speaker's side pass or fail that same criterion?
• What specific, stated difference between the two cases is offered to justify treating them differently, and is that difference relevant to the criterion itself (rather than just who is being judged)?
• Would the speaker accept the same exempting difference as an excuse if the roles of the two parties were swapped?
• Was the exempting distinction announced before the case that benefits from it arose, or was it introduced only after the outcome was known — i.e., does it look tailored to reach a convenient verdict?

Legitimate when: Asymmetric treatment is not fallacious when a genuinely relevant difference between the cases is identified, stated, and would still justify the asymmetry if the parties were reversed — for example, differing intent, differing authority or role, differing consequences, or differing consent. The whole dispute in a real double-standard case almost always comes down to whether the cited difference is a genuinely relevant one or a rationalization invented to fit the desired conclusion.

Ancient: Aristotle articulated the formal principle that like cases must be treated alike: people equal in some normatively relevant respect must be treated equally with regard to that respect, and treating them unequally requires pointing to a real difference in that respect, not merely a difference of who they are. A double standard is the violation of exactly this principle: it treats two relevantly like cases unlike, without identifying the relevant difference the principle requires. — Aristotle, Nicomachean Ethics, Book V.3 (1131a10-b15); Politics, III.9 and III.12

Built from: Wikipedia — Double standard · Wikipedia — Special pleading · Wikipedia — List of fallacies · Internet Encyclopedia of Philosophy — Fallacies · Logically Fallacious — Double Standard · Stanford Encyclopedia of Philosophy — Equality

Ecological Fallacy

also known as: ecological inference fallacy

Just because a whole group has a pattern doesn't mean any one person in it does.

Priya reads a newsletter over coffee: "Zip codes with more gyms per person have lower average rates of heart disease. You just moved to one of those zip codes, Jordan — you'll probably end up healthier just from living there." Jordan isn't so sure: the healthy number for that whole zip code could be coming from a handful of people who run marathons every weekend, while everyone else there, Jordan included, never sets foot in a gym. The neighborhood-wide number doesn't say anything about what happens to Jordan specifically.

The tell: You hear a statistic about a whole place, school, company, or generation, and then someone quietly swaps it in as a fact about one particular person or case from that group — without showing any evidence about that person.

The move: Group G, taken as a whole, exhibits aggregate property or association P (a rate, average, or correlation computed across group-level totals) → therefore a particular member of G has property/relation P.

Why it fails: A statistic computed over aggregated groups describes the composition of those groups, not the joint distribution of traits within any single member. The same set of group averages or group-level correlations is compatible with many different, even opposite, individual-level relationships (Robinson's 1950 reversal is the canonical demonstration), so nothing about a particular person, case, or unit follows from the group-level figure alone without independent individual-level evidence.

• Is the evidence measured at the level of groups (counties, schools, countries, cohorts) while the conclusion is being made about one specific member of one of those groups?
• Could the individual-level relationship actually run in the opposite direction from the group-level one — a Simpson's-paradox-style reversal — while producing the exact same aggregate numbers?
• Has anyone actually checked individual-level records, or is the group average simply being treated as a stand-in for every member of the group?
• Would drawing the group boundaries differently (different counties, different age bands, different time windows) change the aggregate pattern without changing anything about the individuals inside it?

Legitimate when: When the group is known to be homogeneous enough that individual variation around the group mean is negligible, or when the group-to-individual step is backed by actual individual-level data (e.g., a multilevel/hierarchical model that separately estimates within-group effects rather than reading them off the aggregate), the group figure can serve as a defensible estimate for a randomly selected member — but that support has to be demonstrated, not assumed from the aggregate alone.

Built from: Wikipedia - Ecological fallacy · Wikipedia - List of fallacies · Internet Encyclopedia of Philosophy - Fallacies · Stanford Encyclopedia of Philosophy - Fallacies · Logically Fallacious - Ecological Fallacy

Equivocation

also known as: aequivocatio, equivocatio

Saying the same word twice but quietly switching what it means, so the point only seems to make sense.

A group chat is planning a hike. Jordan writes: "Painkillers get rid of headaches. My annoying coworker is a total headache. So I'm bringing painkillers for him tomorrow, ha." It gets a laugh, but if you slow down, "headache" meant an actual physical pain in the first sentence and just meant "an annoying person" in the second. Swap in the real meanings and the joke stops making sense — pills obviously don't fix an annoying coworker.

The tell: You laugh or nod along, then a beat later think "wait, that doesn't actually add up" — the sentence used the same word twice, but it feels like it snuck from one topic to a different one while your back was turned.

The move: A term T occurs at least twice in the argument (e.g., in two premises, or in a premise and the conclusion) carrying sense T1 at one occurrence and a distinct sense T2 at another. The argument's apparent force depends on treating T1 and T2 as if they were the same fixed meaning, so that the syllogism reads as valid on the surface while actually smuggling in a fourth term (quaternio terminorum) where only three are displayed.

Why it fails: The inference is only licensed if the repeated term denotes one and the same thing across every occurrence; that is what lets a middle term connect two premises or a premise connect to the conclusion. Once each occurrence is subscripted for its actual sense, the argument splits into two separate arguments, each missing a premise linking T1 to T2 — the semantic identity the inference relied on was never there, so what looked like one argument was really two half-arguments stitched together at a shared spelling.

• Can the repeated term be replaced by a single, fixed definition at every occurrence without the argument falling apart?
• Give each occurrence of the suspect term a subscript for its sense in context (T1, T2, ...): does any inferential step actually connect two different subscripts rather than the same one?
• If you rewrite the argument using two different words instead of the one ambiguous word, does the conclusion still appear to follow?
• Is the shift in meaning doing real argumentative work here, or is it incidental wordplay that doesn't affect what's being claimed?

Legitimate when: The term is used univocally throughout (one fixed sense everywhere), or a genuine shift in sense is flagged and separately defended rather than smuggled past the audience as if it were the same claim.

Ancient: Aristotle opens his catalogue of language-dependent fallacies with equivocation (homonymy), observing that because there are more things in the world than there are words for them, a single word inevitably ends up covering more than one thing — and a disputant can exploit that gap by sliding from one thing the word covers to another mid-argument, so the conclusion only seems to follow while the two senses stay unseparated. — Aristotle, On Sophistical Refutations

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Equivocation · Internet Encyclopedia of Philosophy — Fallacies

Etymological Fallacy

Knowing where a word came from doesn't tell you what it means today.

Your friend texts the group chat, 'Can't wait for our beach weekend, finally a real vacation!' Another friend replies, 'Not really a vacation though - "vacation" comes from the Latin "vacare," to be empty. If you're planning meals, driving four hours, and checking work email once, your schedule isn't empty, so by definition this isn't a vacation.' Everyone in the chat knows exactly what a vacation is; the word-origin lecture doesn't change that, it just tries to talk them out of their own beach trip.

The tell: It sounds like a fun fact, but it's being used as a trump card to overrule what everyone in the room actually means when they say the word - notice the jump from 'here's where this word came from' to 'so you're using it wrong.'

The move: Word W originally meant (or derives from roots meaning) M1. Therefore M1 is the 'true' or 'correct' meaning of W. Therefore any present claim, argument, or usage that employs W in its current, conventionally established sense M2 is mistaken, illegitimate, or open to correction on that basis alone.

Why it fails: The argument treats a historical fact about a word's origin as if it settled a present fact about the word's meaning, but those are two different questions and the first does not entail the second. Meaning in a living language is fixed by current, conventional usage among competent speakers, not by derivation; words routinely narrow, widen, invert, or otherwise drift from their roots (the same drift that produced M2 out of M1 in the first place), so 'it used to mean X' is simply not evidence for 'it means X.' The move is a species of the genetic fallacy (evaluating a thing by its origin rather than its present content) applied specifically to word-meaning, and it operates as a covert equivocation: the arguer silently substitutes the historical sense for the sense their interlocutor is actually using, then treats the resulting clash as a factual error rather than the switch they engineered.

• Is the claim actually about what the word meant historically (where etymology is real evidence) or about what the word means today (where it isn't)?
• Would competent current speakers recognize the 'true meaning' being urged, or is the arguer trying to legislate a usage nobody actually follows?
• Is the etymological premise even accurate, or is it a folk etymology, a myth, or an invented root?
• If the word's history turned out to be different, would the arguer's substantive point survive? If not, the origin story is doing all the work that evidence should be doing.

Legitimate when: Etymology is genuinely dispositive when the question on the table is itself historical or philological: construing a term of art in a centuries-old statute or contract in the sense its drafters used, translating or glossing an ancient text, or tracing how a word's sense changed over time as an object of study. In those contexts the original meaning is exactly what is being asked about, not a stand-in for what the word means now.

Ancient: In the dialogue, Socrates carries out a long, elaborate survey deriving the 'correct' meanings of Greek names from their components and roots - for instance unpacking 'Astyanax' ('town-lord') to reveal something supposedly true about kingship - as if a name's makeup fixes what it truly signifies. According to the Stanford Encyclopedia of Philosophy's entry on the dialogue, most commentators find this etymological survey riddled with 'blatant mistakes' and 'obviously absurd,' and many read it as Socrates' own parody of the exegetical-etymology method rather than a method he endorses; the dialogue's conclusion is that chasing names back to their roots only recovers the possibly-mistaken opinions of whoever coined them, not the truth about the things themselves. — Plato, Cratylus (394e-421c)

Built from: Wikipedia - Etymological fallacy · Wikipedia - List of fallacies · Internet Encyclopedia of Philosophy - Fallacies · Wikipedia - Genetic fallacy · Stanford Encyclopedia of Philosophy - Fallacies · Stanford Encyclopedia of Philosophy - Plato's Cratylus

Exclusive Premises

Crossing two different things off the same list doesn't tell you anything about how those two things relate to each other.

In the group chat, someone argues: "None of the vegetarians in the office ordered from the taco truck today. And some people who ordered from the taco truck aren't on the softball team. So some people on the softball team must not be vegetarians." It sounds like it clicks into place, but look closer: the taco truck is just a shared thing that two different groups both happen to avoid or ignore in different ways. Nobody ever said one word about softball players and vegetarians actually overlapping or not. The whole claim was built by ruling people out of a third, unrelated group.

The tell: It feels like a fact got smuggled in for free. You can point to what each piece of evidence rules OUT, but you can't point to anything that ever put the two things being compared in the same room together.

The move: A categorical syllogism in which both premises are negative (each an E proposition — 'No M is P' — or an O proposition — 'Some M is not P'). Schematically: No M is P (or Some M is not P); No S is M (or Some S is not M); ∴ [some claimed categorical relation between S and P]. Example instance (EOO-4 form): No M is P. Some P is not S. ∴ Some S is not M.

Why it fails: A negative premise only reports that one class is wholly or partly excluded from the middle term; it never reports how that class relates to the OTHER end term. When both premises are negative, each end term (S and P) is merely pushed away from the middle term M, but nothing in either premise says how S and P stand with respect to EACH OTHER — they could overlap completely, partly, or not at all, and both premises would remain equally true in every case. The middle term's entire logical job is to serve as a shared bridge that positively or negatively links S and P; two premises that only exclude things from that bridge never actually cross it. This is codified as one of the classical rules of the valid categorical syllogism: no valid syllogism can have two negative premises. A Venn diagram in which both premises are satisfied while the conclusion's claimed relation fails to hold can always be constructed, which is what makes the form invalid regardless of the specific terms substituted in.

• Translate both premises into standard categorical form (A/E/I/O) — are both an E or an O proposition (i.e., does each one use 'no' or 'some ... is/are not')?
• Diagram the three classes: after each end term is excluded (wholly or partly) from the middle term, is there any line left in the diagram that actually connects the two end terms to one another?
• Is each premise genuinely negative in logical force, or is one only verbally negative and obversible into an affirmative (e.g., 'No S is non-P' converts to 'All S is P')?
• Swap in a completely unrelated middle term that still makes both premises true — does the 'conclusion' still get asserted with the same confidence? If so, the middle term was never doing any bridging work.

Legitimate when: Never, within standard categorical syllogistic logic — this is a hard structural rule (one of the classical rules of the syllogism), not a defeasible heuristic that sometimes gets overridden by content. The only 'exceptions' are apparent, not real: they occur when a premise that looks negative on its surface is actually affirmative in logical force and can be obverted into an A or I proposition (e.g., 'No contractors are non-licensed' obverts to 'All contractors are licensed'). Once obverted, the argument no longer has two genuinely negative premises and is a different (and possibly valid) form.

Ancient: In his systematic survey of the syllogistic figures, Aristotle shows that when both premises deny rather than affirm a connection to the middle term, no conclusion relating the two outer terms can be drawn: denying the middle term of (or to) both extremes severs each of them from the bridge without ever relating them to one another. — Aristotle, Prior Analytics, Book I

Built from: Wikipedia · Logically Fallacious

Existential Fallacy

Talking as if something definitely exists, just because a rule would apply to it if it did.

A shift manager says in the group chat: "Every employee who breaks the dress code gets written up. Every employee who gets written up loses this month's bonus. So some of us are losing our bonus this month." Everyone panics — but nobody has actually checked whether a single person broke the dress code. Both rules could be perfectly true even if zero people qualify, which means the scary last sentence isn't backed by anything real yet.

The tell: The rules sound solid and official, but you realize no one has actually pointed to a real case — you're being asked to picture consequences for people who might not even exist.

The move: All A are B; All C are A; therefore Some C are B (equivalently, any categorical syllogism mood that combines two universal statements — 'All' or 'No' — as its starting statements but draws a particular 'Some...' conclusion, e.g. the classical moods AAI-1, EAO-1, AEO-4, EAO-4).

Why it fails: In modern (Boolean) categorical logic, universal statements ('All A are B', 'No A are B') carry no existential import: they are true even when the class A is empty, because they only assert a conditional relationship ('if something is A, it is B'), not that any A exists. Particular statements ('Some A are B', 'Some A are not B'), by contrast, do assert that the class in question has at least one actual member. A syllogism that reaches a particular conclusion from two universal starting statements therefore smuggles in an existence claim that neither statement made. If the relevant term picks out an empty class, both starting statements can be true while the conclusion is false, so the move is formally invalid on the modern reading.

• Does the conclusion say 'some' (or otherwise claim a class actually has a member) while both starting statements only say 'all' or 'no'?
• Can you construct a case where the class in question is empty — and check whether the starting statements would still be true even so?
• Which classical syllogistic mood is this (AAI, EAO, AEO, EAO-4, etc.), and is it a 'weakened' mood that is valid only under the traditional Aristotelian existential-import reading?
• Has anyone actually pointed to a real member of the class, or is its existence being assumed purely from what the rule would do to it if it existed?

Legitimate when: When the non-emptiness of the class is independently established — either by explicit stipulation or context showing at least one member actually exists, or by adopting the traditional Aristotelian convention (used before the 19th-century Boolean reformulation) under which universal statements about a subject term presuppose that the term is non-empty. Under that older convention, moods like AAI-1 are legitimate; the fallacy is specifically an artifact of judging such moves by the modern no-existential-import standard.

Built from: Wikipedia — Existential fallacy · Stanford Encyclopedia of Philosophy — The Traditional Square of Opposition · Wikipedia — Existential import · Logically Fallacious — Existential Fallacy

Extended Analogy

Someone compares two different things to the same third thing, each for a different reason, then acts like that makes the two things alike to each other.

In a group chat, Mia says, 'Skipping the gym today is like skipping a work meeting -- you're blowing off a routine you committed to.' Later Jax says, 'Honestly, my cousin skipping his sister's wedding was also like skipping a meeting -- he blew off something he was supposed to show up for.' Then someone pipes up: 'So skipping the gym is basically as bad as skipping a wedding!' But nobody ever showed that skipping a workout and skipping a wedding are alike -- they were each compared to 'a meeting' for completely different reasons.

The tell: You notice the two things being lumped together were never actually held up next to each other -- they were each separately compared to some third thing, for different reasons, and someone just glued the two comparisons together and called it one.

The move: A is likened to a shared anchor B in one respect (A ~r1 B). Separately, C is likened to that same anchor B in a different respect (C ~r2 B), where r1 ≠ r2 and no relevance between r1 and r2 has been shown. The two independent comparisons are then merged through the common label B into a new, unargued claim: A is like C (in whatever respect is currently at issue) — even though A and C were never directly compared to one another.

Why it fails: The move trades on equivocation over the shared middle term B. Being similar to B in respect r1 and being similar to B in respect r2 does not establish that A and C are similar to each other in any respect, let alone the one now being asserted. What actually does the work in each original comparison is the specific respect (r1, r2), not the label 'B' — but the fallacy silently substitutes the label for the respect, so a resemblance between A and B, and an unrelated resemblance between C and B, gets read as a resemblance between A and C.

• Was A ever directly compared to C, or only to the shared anchor B?
• Do the two respects (why A resembles B, and why C resembles B) actually overlap, or are they different properties that just happen to point at the same B?
• Would the person who first said 'A is like B' agree that this commits them to 'A is like C'?
• If you swap out B for a different shared anchor, does the inferred A-is-like-C link survive, or does it evaporate?

Legitimate when: The two separate comparisons share the very respect that is doing the work — r1 and r2 turn out to be the same relevant property (not merely the same label B) — so that A and C, though never directly compared, are both independently shown to possess that property, and transitivity of that specific, argued-for property genuinely holds.

Built from: Stanford Encyclopedia of Philosophy - Fallacies · Logically Fallacious - Extended Analogy · Wikipedia - False analogy · Wikipedia - Argument from analogy · Wikipedia - List of fallacies · Internet Encyclopedia of Philosophy - Fallacies

Fact-to-fiction Fallacy

Someone starts with one real, checkable fact and then rides it, without ever slowing down, straight into a made-up worst-case story — but keeps talking in the same certain voice the whole way, so the story sounds just as solid as the fact did.

A friend texts the group chat: 'The thermostat reading says it hit 91 degrees in the apartment today.' True, verified, nobody disputes it. Same breath: 'Which means the wiring in the walls is degrading, which means we're looking at an electrical fire risk by winter, which means we basically need to break the lease now.' Notice the '91 degrees' part is the only thing anyone actually checked. Everything after it is guesswork stacked on guesswork, but it's delivered in the exact same flat, matter-of-fact tone as the thermostat reading, so it lands like it was also measured.

The tell: You can point to the one number or fact that's actually true, but you can't point to where the rest of the story stopped being checked and started being imagined — it all arrived in the same confident tone, with no seam.

The move: A fact F is established (measured, observed, or well-documented). The arguer extends F through a chain of further claims F -> S1 -> S2 -> ... -> Sn, where each Si is an unmeasured, unmodeled extrapolation rather than a datum, and the chain's plausibility weakens at every step. The terminal claim Sn is then asserted with the same evidentiary confidence as F itself, as though the whole chain were still 'the data talking' rather than speculation stacked on a fact.

Why it fails: The defect is not that reasoning forward from a fact is illegitimate — extrapolation and forecasting are indispensable and often reasonable. The fallacy is a confidence-transfer error: the certainty that properly attaches only to F (the measured premise) is carried, unadjusted, all the way to Sn, even though each intervening step adds its own independent uncertainty that compounds rather than cancels. The arguer never marks the seam between what was actually observed and what was merely imagined forward from it, so the audience has no way to see that the chain's overall reliability is far weaker than its first link. It is a species of weak induction (each extrapolative step is, at best, a probabilistic guess, not an entailment) dressed up rhetorically as a report of fact — the informal-fallacy move is the illicit transfer of the fact's epistemic authority onto the fiction built on top of it.

• Which specific statement in the chain is the actual measured or documented fact, and where exactly does the chain move past it into projection?
• Does the confidence or certainty of the final claim match the weakest, most speculative link in the chain, or only the strength of the original fact?
• If each extrapolative step were separately assigned its own probability, would the compounded probability of the end conclusion still sound as certain as the arguer claims?
• Is the speaker explicitly flagging the extrapolation as a scenario or projection, or is it being narrated in the same voice and tense as the established fact so the two blur together?

Legitimate when: Reasoning forward from an established fact through a modeled chain of inference is legitimate — and not this fallacy — when each step is itself independently justified (by a validated model, a stated base rate, or explicit causal evidence) and the final claim's stated confidence is properly discounted to reflect the uncertainty accumulated along the chain. A climate projection, an epidemiological forecast, or a financial risk model can move from hard data to a speculative future state without committing this fallacy, provided the uncertainty is carried forward honestly and the projection is presented as a projection, not re-asserted as measured fact.

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — List of fallacies · Wikipedia — Slippery slope

Failure to Elucidate

Explaining a confusing thing by using words that are even more confusing, so nobody actually learns anything, it just sounds like they did.

In the family group chat, someone asks why the Wi-Fi keeps dropping. Their tech-savvy cousin replies: "It's probably channel congestion causing DHCP lease renegotiation loops on the mesh backhaul." A few people reply with a thumbs-up emoji and the thread moves on. But nobody in the chat, maybe not even the cousin, could explain what any of those words mean or what to actually do about the Wi-Fi.

The tell: You feel like you just got an answer, but if someone asked you to repeat it back in your own words, you'd have nothing.

The move: Term or phenomenon X is explained (or defined) as Y, where Y is itself less familiar, less intelligible, or more obscure to the intended audience than X was. The explanation has the grammatical shape 'X because/means Y,' but Y cannot be cashed out in terms the audience already understands, so no clarification actually occurs.

Why it fails: A genuine explanation is supposed to move an audience from not understanding X to understanding X by grounding X in something the audience can already independently grasp or check. Here the grounding term Y fails that job: it is no more accessible than X, or less. The audience's confusion is not resolved, only relocated into new-sounding vocabulary, while the surface form of the utterance ('here is why/what X is') creates the appearance that the explanatory debt has been paid.

• Is the offered explanation actually more obscure, to this audience, than the term or claim it was supposed to clarify?
• Could a competent listener restate the explanation in independent terms they already understood before hearing it?
• Does the explanation reduce the listener's uncertainty, or does it just relocate the same uncertainty into new-sounding words?
• Would the explanation still sound informative if it were translated into plain language, or does its authority depend on the jargon itself?

Legitimate when: When the audience actually possesses the technical vocabulary being invoked, so what looks obscure to an outsider is genuinely more precise and independently checkable for that audience. Citing Fourier's law to explain why an oven feels hot is illuminating for a physicist, even though it explains nothing to a layperson standing in the kitchen. Obscurity is audience-relative, not an intrinsic property of using technical or unfamiliar terms.

Ancient: The Skeptic philosopher Sextus Empiricus criticized the atomist Democritus's account of the gods by observing that Democritus tried to clarify something already doubtful (the nature of the divine) by grounding it in something even more doubtful. Sextus's phrase for this move, rendered in Greek as roughly 'the less doubtful by means of the more doubtful,' was later Latinized in the same spirit as the traditional tag obscurum per obscurius (the obscure explained by the more obscure). — Sextus Empiricus, Against the Physicists / Against the Ethicists, I.34, commenting on Democritus's account of the divine

Built from: Wikipedia — Ignotum per ignotius · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies (lists "Obscurum per Obscurius") · Wikipedia — Circular definition

Fake Precision

Piling on decimal points to make a guess sound like a fact.

Your friend is splitting a restaurant bill and announces, 'I already worked it out — you owe $47.83, I owe $52.17.' Nobody checks the math, because a number that lands right down to the penny sounds like it came off a receipt, not out of someone's head a minute ago.

The tell: You never actually saw the measurement, the survey, or the receipt behind it — just a number that sounds too exact to be a guess, so you stop questioning whether it might be one.

The move: A figure E is reported at precision p (a number of significant digits, decimal places, or a narrow interval), where the measurement instrument, sample size, or model actually generating E supports only a much coarser precision q, with q << p. The surplus digits are then treated, implicitly or explicitly, as if they were themselves evidence of accuracy.

Why it fails: Precision (how many digits a figure carries) and accuracy (how close that figure is to the truth) are independent properties, and fake precision trades on collapsing the two. The arithmetic used to produce the extra digits can be flawless — converting 60 mph exactly yields 96.56064 km/h — while the result still misrepresents the underlying uncertainty, because no downstream calculation can manufacture certainty that was not present in the original measurement or sample. The exactness becomes a rhetorical stand-in for evidence the source doesn't actually have.

• What was the actual measurement instrument, sample size, or model behind this figure, and what margin of error does that imply?
• Would the number survive being rounded to the coarsest precision the method can honestly support?
• Is a coarse or rough input being silently carried through a calculation into a falsely exact output (e.g., 'about 60 mph' becoming '96.56064 km/h')?
• Are the extra digits doing persuasive work on their own, independent of any additional evidence being offered?

Legitimate when: The stated precision is actually earned by the instrument, sample size, and propagated uncertainty of the calculation — e.g., a scale accurate to 0.01 g reporting a mass to two decimal places, or a large, well-designed survey reporting a percentage alongside its real margin of error.

Built from: Wikipedia — False precision · Logically Fallacious (Bo Bennett) — Fake Precision / Overprecision

Fallacy of (the) Undistributed Middle

also known as: undistributed middle, non distributio medii

Two things sharing one trait doesn't make them the same kind of thing.

A coworker leans over in the break room and says, 'Everyone on the marketing team carries one of those insulated laptop bags. Dave's got one of those bags. So Dave must be on marketing.' Except half the building carries those bags — IT gave them out as a swag gift two years ago.

The tell: The nagging thought 'wait, tons of other people also do that one thing' — the shared detail doesn't actually narrow anything down, so the leap to 'must be the same group' feels too easy and too confident for how little it's built on.

The move: All M are P (major premise); All S are M (minor premise, or S is M for a singular case); therefore All S are P. The middle term M links the two premises but never appears in the conclusion; the fallacy occurs when M is not distributed (not asserted of every member of M) in either premise, e.g. 'All P are M; All S are M; therefore All S are P.'

Why it fails: A term is 'distributed' when a statement makes a claim about every member of its class. Validity requires the middle term to be distributed at least once, so that both outer terms are pinned to the same members of the middle class rather than to two possibly disjoint slices of it. When neither premise distributes the middle term, both premises only say 'this thing sits somewhere inside M' — they never establish that S's slice of M and P's slice of M overlap, let alone coincide. The conclusion asserting S are P (or vice versa) is a non sequitur: it is easy to construct true premises of this shape with a false conclusion, which is the formal signature of invalidity.

• Restated in standard form, what are the three terms (major, minor, middle), and is the middle term distributed in at least one premise?
• Do the premises actually connect the two outer terms through the middle term, or do they merely place both inside the same middle category?
• Can you name two different subgroups of the middle category — one that fits the first premise, one that fits the second — that don't overlap with each other?
• If you swap in an obviously false pair of outer terms that still fits the same 'both inside M' pattern, does the argument still look convincing? If so, the pattern itself is doing the work, not the specific content.

Legitimate when: The move is legitimate once the middle term is distributed in at least one premise — that is, once one premise makes a universal claim covering the whole of the middle class (a universal affirmative with M as subject, or either term of a universal negative). In that case the two premises are forced to overlap on the same members of M, and a valid categorical mood (e.g. Barbara: All M are P; All S are M; therefore All S are P, with M distributed as subject of the major) goes through.

Ancient: Aristotle's survey of the figures rejects premise combinations that fail to link the extremes through the middle term: for such combinations he exhibits term-triples on which the premises hold while any proposed conclusion fails, which is precisely the counterexample method this record prescribes. The scholastic label 'undistributed middle' (non distributio medii) is a later coinage for the defect Aristotle's system already excludes. — Aristotle, Prior Analytics, Book I (theory of the syllogistic figures)

Built from: Wikipedia — Fallacy of the undistributed middle · Wikipedia — List of fallacies · Stanford Encyclopedia of Philosophy — Fallacies · Logically Fallacious — Undistributed Middle · Wikipedia — Politician's syllogism

Fallacy of Composition

also known as: compositio, composition fallacy

Just because every little piece of something has a certain quality doesn't mean the whole thing has that quality too.

A friend is picking a fantasy sports lineup and says, "I picked the single best player at every position, so this has to be the best team ever." Then the team loses badly, because five ball-hogs who each average forty points a game on their own can't all take the last shot at the same time, and nobody passes. Great on their own, terrible stacked together.

The tell: You hear someone add up a bunch of "each one of these is great" and jump straight to "so the whole pile must be great," and something in you says wait, that's not how it works when you put them all together.

The move: Every part x of whole W has property F (F(x₁), F(x₂), ..., F(xₙ)); therefore the whole W has property F. The move treats F as if it were automatically preserved when parts are aggregated into a composite, without independently establishing that F is the kind of property aggregation preserves.

Why it fails: Whether a property transfers from parts to the whole depends entirely on what that property is, and this is never settled by the mere fact that every part has it. Some properties are aggregate-preserving (every part is made of matter, so the whole is made of matter); many are not, because the whole is a structured arrangement of the parts, not a mere heap, and arrangement itself can create properties absent from any part (emergence, e.g. no single neuron is conscious) or destroy/invert properties every part had (composition effects, e.g. every line item is affordable alone but the sum outstrips revenue; every individual can see better by standing, but not everyone can stand and see better at once). The inference supplies no reason to expect either outcome, so it is invalid unless a separate, property-specific argument for aggregate-preservation is added.

• Is the property F one that survives aggregation (mass, being made of atoms) or one that can vanish, invert, or emerge only at the level of the whole (affordability, being in the minority, being audible, being conscious)?
• Does the argument treat the whole as a mere heap of independent parts, or is the whole actually a structured system where the parts interact, compete for shared resources, or constrain one another?
• If every part were replaced one at a time while holding the others fixed, would F still hold of the whole at every step — or does F depend on how many parts, or which combination of parts, act at once?
• Is there an independent argument that F is aggregate-preserving in this domain, or is the whole's possession of F simply being assumed from the parts' possession of it with no such argument at all?

Legitimate when: The property in question is demonstrably aggregate-preserving in the relevant domain — genuinely additive or extensive properties like total mass, total volume, or total cost transfer straightforwardly from parts to whole because that is what such properties mean; the inference is safe only when this closure has been independently shown, not merely assumed by analogy with properties that do transfer.

Ancient: Aristotle lists composition among the language-dependent fallacies and illustrates it with sentences whose truth flips depending on whether words are taken separately or combined, such as the claim that a man can walk while sitting: true if 'able to walk' and 'sitting' are kept apart as separate facts about him, false if compressed into the single claim 'walking-while-seated.' Note this is Aristotle's own sense of the fallacy — a scope ambiguity about how words combine in a sentence — which historians of logic distinguish from the modern parts-to-whole inference described above; the two share a name and a family resemblance but are not the identical mistake. — Aristotle, On Sophistical Refutations, ch. 4

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Fallacy of composition · Internet Encyclopedia of Philosophy — Fallacies

Fallacy of Division

also known as: divisio, division fallacy

Assuming that because a group or a whole thing has some trait, every piece or member of it must have that same trait too.

A friend says, "Your book club is so smart — you guys read like three novels a month and always have deep discussions." Then she meets your friend Marco from the club at a party and says, "Oh, so you must be a huge reader and super into philosophy, right?" Marco actually just shows up for the snacks and rarely finishes the book. The club as a whole having a reputation for being well-read doesn't mean every single person in it is.

The tell: You notice someone is judging one specific person, item, or part based only on a label that describes the whole group or object — and you want to say "wait, you don't actually know that about him, you just know it about the group." It feels like they skipped a step: they never actually checked the individual, they just borrowed the group's reputation and pasted it onto him.

The move: The whole W (or the group G) has property F. X is a part of W (or a member of G). Therefore, X has property F.

Why it fails: The inference silently assumes that F is a distributive property — one that automatically passes from a whole or group to each of its parts or members — when in fact F may be a collective (or emergent) property that belongs to the whole only in virtue of how the parts are arranged, combined, or aggregated together. A team can be undefeated without every player being individually excellent; a chamber's low approval rating or gridlock describes the body's output as a collective, not a trait any single member need share. Because the argument never establishes that F actually distributes in this domain, the move from 'the whole has F' to 'the part has F' is a gap the premises do not license — the part is simply assumed to inherit whatever is true of the aggregate.

• Is the property being attributed collective (true of the group or whole only as an aggregate) or distributive (true of each member taken individually)?
• Does the whole possess this property only because of how its parts are arranged, combined, or interact — an emergent feature that no single part need share?
• Can you name one member of the whole that clearly lacks the property while the whole still has it?
• If you swapped out this one part for a different one, would the property of the whole change at all — or is the whole's property independent of which particular part you're looking at?

Legitimate when: The move is legitimate when the property really is distributive in that domain — genuinely and demonstrably true of each part simply because it is true of the whole (for example, 'this whole shipment is contaminated with substance X' licenses 'therefore this crate from the shipment is contaminated,' if contamination was established to be uniform throughout, or 'every part of this wall is painted blue' trivially licenses that any given brick-sized patch is blue).

Ancient: Aristotle's own example of division is about numbers, not groups and members: he shows how one could argue that five is both even and odd by wrongly dividing it into two and three — two is even, three is odd, so (improperly dividing the compound) five is said to be both. This is a syntactic move about how a composite term gets split apart, distinct from the modern collective-to-member inference that later logicians (from Whately onward) came to call the fallacy of division. — Aristotle, Sophistical Refutations, ch. 4 (fallacies dependent on language: division)

Built from: Wikipedia — Fallacy of division · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies

Fallacy of Every and All

Everybody having an answer doesn't mean everybody has the same answer.

In the office group chat someone writes: "Every single person on this team has a productivity hack that works for them. So there must be one productivity hack out there that works for everyone — we just need to find it and make it the team standard." But Dana swears by silence and noise-cancelling headphones, Priya needs a chatty open room to think out loud, and Marcus gets nothing done without a strict hour-by-hour schedule. Each person really does have something that works — it's just not the same something.

The tell: It sounds like a tidy discovery — "oh, so there's one big answer for everybody!" — but nobody ever actually named that one shared answer; it was just assumed to exist because each person, individually, had some answer of their own.

The move: For every A, there is some B such that A relates to B (∀x ∃y R(x,y)). Therefore, there is some one B such that every A relates to it (∃y ∀x R(x,y)). The conclusion illegitimately pulls the existential quantifier outside the universal one — letting one shared witness stand in for what was really a different witness for each case.

Why it fails: The premise only guarantees that a B exists for each A separately — the identity of B is allowed to change from case to case. The conclusion asserts something much stronger: one single, fixed B that works for every A at once. Nothing in "for each A there is some B" rules out each A having its own distinct B, so the argument gives no ground for collapsing the many possible B's into one common B. The reverse move (from "one B for all A" to "each A has some B") is fine — the problem is only in this direction.

• In the premise, does the thing satisfying "some" get to be a different one for each case, or is it already fixed as a single item?
• Does the order of the quantifiers actually change between premise and conclusion — does "for every X there is some Y" become "there is some Y for every X"?
• Can you build a scenario where every individual case has its own separate answer, so that no single shared answer covers them all?
• Is the word doing the work here "every," "all," or "any" — and is its scope (how far it reaches over the rest of the sentence) ambiguous in the wording used?

Legitimate when: The move is legitimate only when a single common item has been independently shown to satisfy the relation for every case — that is, when ∃y∀x R(x,y) is established on its own terms, not merely inferred backward from ∀x∃y R(x,y). It is also fine to reason in the other direction: from "one thing works for everyone" it does validly follow that "everyone has something that works for them."

Ancient: Aristotle opens by observing that every action of ours aims at some end or good. From this, the argument is commonly read as concluding that there is therefore some one common end — a single supreme good — toward which all our actions aim, which he identifies with happiness. The IEP flags this as a textbook case of the fallacy: each action having its own end does not by itself establish that one single end is shared by all of them. — Aristotle, opening argument of the Nicomachean Ethics, Book I (as read by the Internet Encyclopedia of Philosophy's "Fallacy of Every and All" entry)

Built from: Internet Encyclopedia of Philosophy — Fallacies · Wikipedia — Quantifier shift · Stanford Encyclopedia of Philosophy — Fallacies

Fallacy of Four Terms

also known as: quaternio terminorum, four-term fallacy

It's when an argument reuses the same word twice as if it means one thing, but the word is secretly doing two different jobs, so the two halves of the argument never actually touch.

In the group chat, Mateo posts: 'Free stuff is always worth grabbing. This gym's "free" trial just needs your card number and auto-bills you after a week. So you should definitely grab it.' The word 'free' is pulling double duty - in the first line it means 'costs you nothing,' in the second it means 'advertised with the word free.' Swap in the real meanings and the two sentences don't add up to anything.

The tell: The two sentences feel glued together because they share a word, but the moment you try to say out loud exactly how one leads to the other, the glue turns out to be nothing but the repeated word itself.

The move: A syllogism displays the standard three-term skeleton (major term, minor term, and a middle term meant to link the two premises) but in fact smuggles in a fourth term: one word or phrase - almost always the middle term - is used in two different senses across its two occurrences, so what looks like one shared term is really two unrelated ones.

Why it fails: A categorical syllogism is only valid if the middle term is genuinely identical - same referent, same sense, same scope - in both premises, because that identity is the sole thing that licenses connecting the major and minor terms in the conclusion. If the two occurrences of the 'shared' term actually pick out different things, the argument silently contains four terms rather than three, no premise is logically linked to the other through a common middle, and the conclusion does not follow from what has actually been asserted - it only follows from the surface repetition of a word.

• List every term exactly as used (not as spelled) - are there really only three, and does the middle term mean the same thing both times it occurs?
• If you gave the two occurrences of the suspect term separate labels (e.g. 'bank-1' and 'bank-2'), would the premises still share any term at all?
• Does the repeated term shift between a literal/technical sense and a loose/figurative one, or between two different scopes of the same word (e.g. 'nothing' meaning 'no thing whatsoever' vs. 'nothing' meaning 'a thing of negligible value')?
• Which single premise would have to be dropped, and which sense of the word abandoned, before the argument could be rewritten as a genuine three-term syllogism?

Legitimate when: The move is simply correct syllogistic reasoning whenever the recurring term is truly univocal - the exact same concept, sense, and scope, in both premises. An ordinary, cleanly three-termed categorical syllogism is not a trick at all; the fallacy exists only in the gap between apparent and actual term-identity.

Ancient: Among Aristotle's language-dependent fallacies are arguments that trade on a single word carrying two different senses across the reasoning; read as a syllogism, such a double-sensed middle term counts as two terms rather than one, a defect later scholastic logicians formalized as quaternio terminorum. — Aristotle, Sophistical Refutations (fallacies of homonymy and equivocation)

Built from: Wikipedia - Fallacy of four terms · Wikipedia - List of fallacies · Internet Encyclopedia of Philosophy - Fallacies

Fallacy of Opposition

Hearing that 'all of them' isn't true and jumping straight to 'none of them' — when the honest answer is almost always 'some, but not all'.

At book club, Maya says everyone loved the ending. Priya sets down her glass: "Not everyone — Dev thought it was a cheat, and honestly so did I." Maya throws up her hands: "Great, so nobody liked it, we wasted a whole month on that book." Three people at the table did like it, and one of them is still holding the book. Maya has gone from 'not all' to 'none' in a single breath, and the room lets her, because the two sentences sound like the same objection.

The tell: You feel yourself shoved out of a moderate position you actually hold and into an extreme one you never said. You raised one exception and it comes back to you as a blanket denial, and now you are defending a claim that isn't yours.

The move: Two categorical statements about the same subject S and predicate P stand in one of the four relations of opposition: contradictory (A/O and E/I), contrary (A/E), subcontrary (I/O), or subalternation (A to I, E to O). The arguer establishes the truth value of one member of a pair and then transfers it to the other as though every relation of opposition behaved like contradiction — most often moving from the falsity of 'All S are P' to the truth of 'No S are P', or from the falsity of 'Some S are P' to the truth of 'All S are not-P'. What is licensed is only exclusion; what is taken is exhaustion.

Why it fails: Only contradictories divide the space of cases exhaustively: exactly one of 'All S are P' and 'Some S are not P' must hold, so knocking down one does erect the other. Contraries merely cannot both be true — 'All the reviews were positive' and 'None of the reviews were positive' can both be false at once, and typically are, because the real state of affairs is mixed. Subcontraries are the mirror image: 'Some are' and 'Some are not' cannot both be false, but they can happily both be true, so establishing one settles nothing about the other. The argument runs a rule that holds for one pair across all four pairs, and the middle ground it silently deletes — some but not all — is almost always where the truth lives. The result is a manufactured either/or produced by a misread of the diagram rather than by anything about the subject matter.

• Write both statements out in plain 'All / No / Some / Some are not' form. Which pair is actually on the table?
• Can you describe one concrete situation in which both statements are false at the same time? If you can, the falsity of one does not establish the other.
• Does the argument move from 'not all' to 'none', or from 'not none' to 'all'? Those two steps are the commonest form of this error.
• Replace the denied statement with its exact contradictory — 'at least one is not' rather than 'none is' — and re-read the argument. Does the conclusion still follow?

Legitimate when: The transfer is legitimate when the pair really are contradictories: from the falsity of 'All S are P' you may infer 'Some S are not P', and from the falsity of 'Some S are P' you may infer 'No S are P'. Traditional logic also licenses subalternation — 'All S are P' to 'Some S are P' — provided the subject class is known to be non-empty, which is the assumption modern logic declines to make for free.

Ancient: Aristotle sets out the opposition between statements, separating contradictory pairs from contrary ones. Contradictories divide truth and falsity between them exactly; contraries such as 'every man is just' and 'no man is just' cannot both be true, but he is explicit that they can both be false. The distinction he draws there is precisely the one this error collapses. — Aristotle, De Interpretatione 6–7

Built from: Logically Fallacious — Fallacy of Opposition · Stanford Encyclopedia of Philosophy — The Traditional Square of Opposition · Wikipedia — Square of opposition

False Attribution

Propping up your point with a quote or a study that either doesn't exist, doesn't say that, or comes from someone with no standing to say it.

In the family group chat, Aunt Deb posts a graphic: "If the bee disappeared, man would have only four years left to live. — Albert Einstein." Her nephew Sam writes back, "Where's that from? Which letter, which year?" Deb replies with a shrug emoji and "It's Einstein, Sam." Nobody has ever found that line anywhere in Einstein's writing, and the whole force of the post came from the name at the bottom rather than from anything anyone in the chat knows about bees.

The tell: The name lands before the content does, and you notice you're suddenly being asked to argue with a famous stranger instead of with the person in front of you. When you ask where it came from, the request itself gets treated as rude — which is the strongest sign that nobody has ever checked.

The move: An arguer A advances claim C and supports it by attributing an endorsement of C to a source S — a person, study, book, or institution. The support offered is the attribution itself. The attribution is defective in one of four ways: S does not exist or cannot be identified; S exists but never said what is reported; S said it but has no competence on the question; or S has a stake in the answer that the audience is not told about. In every case the weight the argument carries is borrowed against a source that cannot in fact bear it.

Why it fails: Testimony supports a claim only through a chain: a competent, disinterested source examined the matter, said something about it, and the report of what they said is accurate. Break any link and the chain transmits nothing — a fabricated quotation is not weak evidence, it is no evidence, since there was never an examination to inherit. An unidentifiable source is worse than useless because it is also uncheckable: naming 'a study out of Sweden' converts a testable claim into one the audience cannot investigate, which shifts the whole cost of verification onto the listener and leaves the speaker's assertion untouched no matter what. And when the source is real but unqualified, the endorsement is a fact about a person's opinions rather than about the subject. Delete the attribution and inspect the residue: if nothing remains, the name was doing all the work, and the name was hollow.

• Can the source be named exactly — person, work, date — and can the passage be located?
• Read in full and in context, does the source actually say what is being attributed to it?
• Is the source qualified on this specific question, and does the source stand to gain from the answer?
• Strike the source's name out of the argument entirely. Is there anything left that supports the claim?

Legitimate when: Citing testimony is the ordinary and unavoidable way most knowledge travels. The appeal is sound when the source is genuinely identified, accurately reported, competent on the question at hand, and not concealing an interest in the outcome — a named study whose method the audience can inspect, an expert speaking within her field, a witness reporting what she saw. The test is not whether a name is invoked but whether the audience can follow the chain back and check it.

Built from: Wikipedia — False attribution · Logically Fallacious — False Attribution · Stanford Encyclopedia of Philosophy — Fallacies

False Conversion

Flipping 'all X are Y' around into 'all Y are X', as though the two groups had to be exactly the same set of things.

At the dog park, Vera says every dog that got sick last week had eaten the new liver treats. Marcus, who buys those treats by the case, goes pale: "So every dog on those treats is going to get sick." Vera never said that. There are forty dogs at the park on those treats and four of them got sick; the sentence she said points from the sick dogs to the treats, and Marcus has just read it backwards, pointing from the treats to the sickness.

The tell: You agree with the sentence someone just said and somehow feel cornered by what came out of it. The words are the same words in the other order, and the whole weight of the argument is sitting on the swap rather than on anything anybody actually established.

The move: Conversion is the immediate inference that swaps the subject and predicate terms of a categorical statement. From 'All F are G' the arguer draws 'All G are F'; or from 'Some F are not G' the arguer draws 'Some G are not F'. These are the two forms that do not convert. The universal negative 'No F are G' and the particular affirmative 'Some F are G' do convert simply and without loss; the universal affirmative converts only per accidens, weakening to 'Some G are F', and the particular negative does not convert at all.

Why it fails: 'All F are G' asserts containment, not coincidence: it places the F's inside the G's and says nothing whatever about how much of G lies outside F. Swapping the terms replaces that containment claim with an identity claim, which is strictly stronger, so the move manufactures information out of nothing. A counterexample is always available and always the same shape — all dogs are mammals, and the cats are the mammals that are not dogs. The particular negative fails for a related reason: 'Some mammals are not cats' is true while 'Some cats are not mammals' is false, because 'some F are not G' locates a gap on the F side of the relation and gives no licence to look for one on the G side. What makes the error hard to catch in speech is that the converted sentence uses the same two nouns and often sounds like a restatement rather than a new and much bolder claim.

• Rewrite both statements as 'All / No / Some / Some ... are not ...' and check whether the subject term and the predicate term have traded places.
• Which of the four forms is being flipped? 'No F are G' and 'Some F are G' may be flipped freely; 'All F are G' and 'Some F are not G' may not.
• Try to name one G that is not an F. If you can name one and the original statement still stands, the flip has failed.
• Say the converted sentence out loud in the speaker's own words and ask whether they will endorse it as stated — 'and therefore all G are F'.

Legitimate when: Simple conversion is a legitimate immediate inference for the universal negative and the particular affirmative: 'No politicians are saints' really does give you 'No saints are politicians', and 'Some teachers are cyclists' really does give you 'Some cyclists are teachers'. The universal affirmative converts per accidens to a particular — 'All F are G' to 'Some G are F' — provided F is known to have at least one member. And the full swap is legitimate whenever the original claim was in fact a definition or a biconditional, an 'all and only' rather than an 'all'.

Ancient: Near the opening of the Prior Analytics Aristotle works out which statements convert and how. The universal negative and the particular affirmative convert simply; the universal affirmative converts only to a particular, 'if every B is A, then some A is B'; the particular negative he shows does not convert at all. These are exactly the rules whose violation constitutes false conversion, and later logicians named the error against them. — Aristotle, Prior Analytics I.2

Built from: The Fallacy Files — Illicit Conversion · Stanford Encyclopedia of Philosophy — The Traditional Square of Opposition · Wikipedia — Square of opposition

False Dilemma

also known as: false dichotomy, bifurcation fallacy

It's when someone hands you only two choices — usually one reasonable-sounding and one terrible — and hides the fact that other choices exist.

Your roommate says, 'Either we get a dog, or we're the kind of people who don't care about animals.' You never said you didn't want a dog. Maybe you're allergic, maybe your apartment is tiny, maybe you'd rather get a cat next year. None of that fits into either box you were just handed — but the way the sentence is built, saying 'no' to the dog suddenly makes you sound heartless.

The tell: You feel boxed in — like you're being hurried to pick from a menu with only two items on it, and the thing you'd actually say doesn't fit on either side. That cornered feeling is usually the first sign someone erased the middle ground on purpose.

The move: Presented as: (P ∨ Q), ¬P, ∴ Q — a disjunctive syllogism that is structurally fine. The defect sits one level up, in the unstated premise that P and Q are the only options and cannot both hold or both fail (i.e., that P ∨ Q is exhaustive and exclusive). Strip that assumption out and the argument is just P-or-Q with no support for why it must be P or Q at all.

Why it fails: The reasoning step is not broken; the menu is. The arguer treats two contraries (claims that could both be false, or that could both be true, or that leave room for a third) as if they were contradictories (claims where exactly one must be true). Once a genuine third option, a middle ground, or a point on a spectrum between P and Q is produced, the forced choice collapses — nothing about the logic itself was ever in question.

• Are the two options actually mutually exclusive and jointly exhaustive, or could both be true, both be false, or something else entirely be true?
• What third option, middle position, or 'none of the above' has been left off the menu?
• Is the exhaustiveness of the two options actually argued for anywhere, or just asserted by the phrasing ('either... or...')?
• Are P and Q really two discrete boxes, or are they two ends of a continuum being carved into an artificial binary?

Legitimate when: When the two options genuinely are exhaustive and exclusive — a real logical complement (something either happened or it didn't), a law or rule that in fact only allows two outcomes, or a case where the arguer has independently proven that no third option exists. In those cases the same P ∨ Q, ¬P, ∴ Q form is simply sound disjunctive syllogism, not a fallacy.

Built from: Wikipedia — False dilemma · Internet Encyclopedia of Philosophy — Fallacies · Stanford Encyclopedia of Philosophy — Fallacies

False Effect

Deciding a warning is nonsense because the scary outcome it named never showed up — or deciding a cause is real because the outcome did.

Driving back from practice, Dad tells his son that listening to music that loud will leave him deaf by thirty. The son doesn't even look up: "Uncle Ray blasted music his whole life and he's fifty-two and hears fine. So headphones don't hurt your ears, drop it." Dad overstated the outcome, no question. But 'you won't be deaf at thirty' and 'nothing is happening to your hearing' are two different statements, and the son has used the first to close the door on the second.

The tell: You made a specific warning, someone found one hole in the specific part, and now the entire subject is closed. It feels like losing a much bigger argument than the one you were actually having — the overreach got refuted and took the real concern down with it.

The move: A causal claim is on the table: cause C produces effect E. Rather than examining the evidence for the link, a second party evaluates E on its own — finding that the alleged effect did not occur, or is absurd, or did occur — and transfers that verdict back to C. In the negative direction: E is false, therefore C is false, harmless, or not worth acting on. In the positive direction: E is real, therefore C is real, correct, or vindicated. The truth value of the effect is routed to the cause across a bridge that was never built.

Why it fails: The whole transfer depends on the causal link, and the causal link is precisely what nobody established. If C and E were never shown to be connected, then E's status carries no information about C at all — you have learned something about the world and nothing about the claim. Even where the link is real, the negative direction proves far less than it appears to: refuting E refutes the pairing 'C causes E', which leaves entirely open whether C is harmful, real, or worth avoiding for some other reason. Sitting a foot from the television does not cause blindness, and that does not make it good for a child's eyes. The positive direction is worse, because effects are almost never the product of a single candidate cause: finding that E occurred licenses at most 'something produced E', and the argument quietly reads it as 'C did'.

• Write the claim out in the form 'X causes Y'. Is any evidence given for the link itself, or is the link only assumed by both sides?
• Is the argument judging X by what is known about Y, rather than by any evidence about X?
• Suppose the specific outcome Y is exactly as false as claimed. Does the case for or against X have any support left that does not run through Y?
• Could Y have arisen from causes other than X, or could X have effects other than Y, that the argument never mentions?

Legitimate when: The transfer is legitimate when the causal claim is genuinely established and E is a necessary consequence of C: if a disease always produces a particular marker and the marker is absent, the disease is ruled out, which is how diagnostic testing works every day. The condition is that the connection has been demonstrated independently and that the direction of the inference respects it — a missing necessary effect refutes the cause, a present effect merely leaves the cause among the candidates.

Built from: Logically Fallacious — False Effect · Wikipedia — Questionable cause

False Equivalence

Treating two things as the same because they have one thing in common, when one of them is vastly bigger, worse, or more likely than the other.

On the sideline at the under-11s, someone brings up that Coach Dana spent the second half calling her own players useless — again, third week running. Another parent shrugs: "Ellis snapped at the ref back in April. Both of them have a temper. It's a wash, honestly." One coach lost his patience with an adult once; the other has ten-year-olds crying in the car on the way home. Putting both under 'has a temper' makes them one thing, and the conversation about Dana quietly ends.

The tell: You've been maneuvered into defending something you never brought up. Someone has found a category roomy enough to hold your complaint and a much smaller thing, and now you're on the hook for both — and pointing out the difference in size starts to sound like special pleading.

The move: Two items X and Y both possess some property P. From shared possession of P the arguer moves to comparable standing — same kind, same magnitude, same moral weight — and then draws a conclusion that depends on that standing: neither is worse than the other, condemning X commits you to condemning Y, the question between them is a wash. The property P invoked is almost always a determinable ('has legal trouble', 'is a weapon', 'made a mistake', 'is a substance people take') whose determinate values differ by orders of magnitude, and it is the determinate value the conclusion actually turns on.

Why it fails: Sharing a property is not sharing a degree of it, and equivalence is a claim about degree. The argument works by ascending to a category broad enough to contain both items and then descending as though the category had settled the comparison, which drops exactly the quantity that made the two cases different in the first place. A pint of motor oil in a driveway and two hundred million gallons in the Gulf are both spills; the classification is correct and it decides nothing. There is a second cost beyond the arithmetic: because equivalence claims are usually deployed to neutralise a judgement rather than to establish one, the move does not merely fail to support its conclusion, it removes the audience's ability to rank anything at all. The test is nearly mechanical — restate the shared property with its magnitudes attached, and if the equivalence evaporates, it was never carried by the property.

• Name in one phrase the exact property the two things are said to share.
• Attach a number or a magnitude to that property for each side. Do the two land anywhere near each other?
• Is the shared property the one the conclusion depends on, or a broader label that merely contains it?
• Swap the two sides and re-run the claim. Would the speaker accept the equivalence in that direction too?

Legitimate when: The comparison is sound when the shared property is held to a genuinely comparable degree and is the property the conclusion turns on — two candidates who each took money from the same donor at similar scale, two employees who each missed the same number of deadlines. Consistency arguments of the form 'you condemned that, so condemn this' are legitimate whenever the cases really do match in the relevant respect and amount; what makes them fail is a mismatch of magnitude concealed by a shared name.

Built from: Wikipedia — False equivalence · Logically Fallacious — False Equivalence · Wikipedia — False balance · Stanford Encyclopedia of Philosophy — Fallacies

Fantasy Projection

Describing a scene you have only ever pictured in your head so completely that everyone starts treating it as something that really happened.

Over coffee, Nadia tells her friend what the lunch table at work is like when she isn't there. "Renata does the voice — my voice — and Tom does that laugh, and then they all go quiet when I come back from the printer." She has never been at that table while it happened, and nobody has told her about it. By the third telling her friend is asking what she's going to do about Renata, and neither of them has stopped to ask how Nadia knows.

The tell: You find yourself arguing about the details of a scene instead of about whether the scene ever took place. It is told like a memory, with the timing and the tone of voice included, but nobody is actually claiming to have been in the room — and asking who saw it feels weirdly aggressive.

The move: A speaker A holds a vivid inner representation X — a pictured scene, a felt certainty, an imagined motive behind someone else's behaviour. A does not offer X as a hypothesis but asserts it as shared fact, and the warrant supplied is the character of the representation itself: its detail, its emotional charge, its coherence. The audience is then required to reason from X, so X enters the discussion as a presupposition rather than as a claim under test.

Why it fails: Vividness is a fact about how well the imagination is working, not about how the world is arranged, and it is precisely non-discriminating: an invented scene and a witnessed one can be equally detailed, so detail cannot separate them, which is the only job evidence has. Because the scene arrives as background rather than as an assertion, the ordinary machinery of objection never engages — the audience finds itself arguing about what follows from the scene instead of whether the scene occurred, and disagreeing sounds like calling the speaker a liar rather than asking for a source. And since nothing observable was ever cited, nothing observable can count against it; the claim is insulated from correction by the manner of its delivery. Where subjective experience is genuinely the subject ('I felt humiliated'), the report is legitimate; the error is generalising the inner scene into a claim about other people's conduct, motives, or the world at large.

• What could someone other than the speaker have observed that supports this scene?
• Is the sheer amount of detail — what people said, how they looked, what they were thinking — standing in for any actual report from anyone?
• Was the speaker present? If not, what is the chain from an actual witness to this description?
• Relabel the scene 'here is how I picture it' and read the argument again. Does anything still follow?

Legitimate when: Reporting your own experience is legitimate and often the only evidence available: 'that meeting felt hostile to me' is a claim about a real thing the speaker has privileged access to. Imagined scenarios are also legitimate as explicit hypotheses, thought experiments, or illustrations, where nothing is inferred about what actually happened. The line is crossed when an unwitnessed scene is asserted as fact about other people, or when a private experience is generalised into a claim about how the world is.

Built from: Logically Fallacious — Fantasy Projection · Wikipedia — Wishful thinking

Far-Fetched Hypothesis

Reaching for a wild explanation before anyone has checked whether the boring one fits.

Cal comes back to the basement laundry room to find his wet clothes stacked on top of the machine. By the time he's back upstairs he has it worked out: the building manager has a master key, the manager has been annoyed since the complaint about the radiator, and this is retaliation designed to look like nothing. His neighbour on the fourth floor had a wash to do at eleven at night and a dryer to catch. Nobody has asked her.

The tell: The story is more interesting than the thing it is explaining, and you catch yourself impressed before you are convinced. It also tends to arrive complete — motive, method, and timing all worked out — which is a strange thing for a true account to be, five minutes in.

The move: Some evidence E stands in need of explanation. A range of hypotheses H1...Hn would account for E, including several mundane ones. The arguer selects a hypothesis with a low prior probability and substantial unobserved machinery — a coordinated scheme, an unprecedented event, an entity nobody has detected — and advances it as the explanation of E, without comparing it against the ordinary rivals and often with the assertion that nothing else fits.

Why it fails: Explanations are not supported by fitting the evidence, because indefinitely many explanations fit any body of evidence; they are supported by fitting it better than the alternatives, which means the comparison is not an optional refinement but the whole of the support. Skipping it leaves the favoured hypothesis untested against anything at all. The exotic option is also the one that can least afford to skip it: each additional unobserved component it requires — the accomplices, the secrecy, the mechanism no one has measured — is a further way for it to be false, so it starts from a lower prior and needs more evidence than a mundane rival, not less. The characteristic move, 'nothing else explains it', is usually not a fact about the world but a report of how long the speaker looked; and even a genuinely exhaustive search only ever establishes the best of the hypotheses actually considered, which is the standing worry about inference to the best explanation.

• Name two ordinary explanations that would fit the same evidence. Did the argument address either of them?
• What does the favoured explanation require to be true that nobody has observed — a group keeping a secret, an unprecedented failure, an undetected mechanism?
• Is the case for it positive evidence, or is it 'nothing else could explain this'?
• What single observation would come out differently under the exotic story than under the ordinary one, and has anyone actually checked it?

Legitimate when: Surprising explanations are sometimes correct, and the standard is comparison rather than familiarity. The reasoning is sound when the mundane rivals have genuinely been examined and ruled out on evidence, when the unusual hypothesis makes a distinctive prediction that is then checked, and when the prior improbability is acknowledged and paid for with correspondingly strong evidence. Continental drift, ulcers caused by bacteria, and the asteroid impact at the end of the Cretaceous all began as far-fetched hypotheses and were established in exactly that way.

Built from: Logically Fallacious — Far-Fetched Hypothesis · Stanford Encyclopedia of Philosophy — Abduction (Inference to the Best Explanation)

Faulty Comparison

Putting two things side by side on a measurement that doesn't mean the same thing for both, then announcing a winner.

Ade holds up a cereal box in the aisle: "Forty per cent less sugar than the leading brand." Jules turns both boxes over. The leading brand counts a serving as fifty-five grams; this one counts thirty. Per bowl — which is what anyone actually pours — they come out within a gram of each other. The claim on the front is perfectly true and tells you nothing, because the two numbers were never counting the same thing.

The tell: The number is true and the point still feels crooked. You want to ask 'compared to what, measured how' and can't quite say why, because nothing you've been told is false — it's what's sitting underneath the two figures that doesn't match.

The move: Two items X and Y are set side by side on a dimension D and a verdict is drawn — X is safer, healthier, more successful, better run. The comparison fails in one of two ways. Either D is a dimension on which the two were never in competition, so a true statement about X's standing on D is made to answer a question it does not address; or the two measurements on D were not produced on the same basis — different denominators, time windows, definitions, populations, or measurement conditions — so the gap between the figures reports the gap between the bases.

Why it fails: A comparison transmits information only when the two measurements differ in the thing being compared and agree in everything else that could move the number. When the bases differ, that condition is broken at the source: one state's raw job count set against a state a fifth its size measures population, not policy; this year's twelve months set against last year's nine measures the calendar. Normalisation is the diagnostic precisely because it isolates the variable of interest, and a comparison that reverses under normalisation was never about what it claimed to be about. The other failure is subtler because every individual statement in it is true: broccoli really does contain less fat than a chocolate bar, and the sentence is still worthless, because nobody was choosing between them and fat was never the dimension on which the choice turned. In both forms the word 'than' does the deceiving, since it implies a shared scale that the two figures do not share.

• Are both figures measured the same way — same definition, same time window, same population, same denominator?
• Normalise them (per person, per hour, per dollar, per serving). Does the comparison survive, or does it reverse?
• Is the dimension being compared the one the decision actually turns on, or a flattering side-dimension chosen after the fact?
• Add a third item to the ranking. Does the ordering still make sense, or does the scale fall apart?

Legitimate when: Comparison is the ordinary way quantities are made meaningful, and it is sound when both sides are measured on the same basis and the dimension is the one the decision turns on: two cars' fuel consumption on the same test cycle, crime rates per hundred thousand across cities of different sizes, two hospitals' outcomes adjusted for how sick their patients were on arrival. Raw totals are fine too, when the total itself is what matters — a budget has to be paid in whole dollars, not per capita ones.

Built from: Logically Fallacious — Faulty Comparison · Wikipedia — False equivalence · Stanford Encyclopedia of Philosophy — Fallacies

Gadarene Swine Fallacy

Deciding that the one person out of step must be the one who is lost, purely because everyone else is moving together.

Two hours into a night shift, Bea flags that the med-timing sheet has the second dose landing inside the four-hour window and she doesn't think that's right. The charge nurse doesn't look at the sheet. "Bea. Six floors run this protocol and nobody else has ever raised it. If you're the only one, it's you." Maybe it is her. But nobody has read the sheet yet, and the answer to whether the window is right is on the sheet, not in the head count.

The tell: The reply is about your position in the room rather than about what you said. You get asked to explain why you are the odd one out instead of being asked to explain your reason, and you notice that nobody has actually looked at the thing you pointed at.

The move: A group G is unanimous, or nearly so, in believing or doing P. An individual D is out of step. The arguer concludes that D is mistaken, or must fall into line, and the whole of the support offered is the fact of G's cohesion — that D is the only one, that everybody else manages, that thirty-one against one settles it. The property invoked is a property of the group's formation; the conclusion is about the direction the formation is heading.

Why it fails: Cohesion has causes other than correctness, and most of them operate whether the heading is right or wrong: conformity pressure, shared incentives, a common information source, and plain copying all produce a tight formation. Where the members are watching each other rather than the question, agreement is derivative — a hundred people repeating one report constitute one piece of evidence, not a hundred, and the argument cannot tell the two situations apart because it counts heads rather than independent examinations. The deeper defect is that the argument is structurally blind in exactly the case where it is most needed: a group that has collectively gone wrong presents the identical evidence, unanimity, as a group that has collectively gone right. And nothing about D's actual reason has been examined; the reply is about D's position in the room rather than about the matter under dispute, which makes it a relevance failure in the ad populum family rather than a weak generalisation.

• Apart from the fact that everyone else is doing it, what reason is offered that the group is right?
• Did the members reach this independently, or are they mostly copying one source and each other?
• Could the group's agreement be fully explained by pressure, habit, or shared incentive rather than by evidence?
• Has anyone engaged the dissenter's actual reason, or only the fact that they are dissenting?

Legitimate when: Agreement is real evidence when it is independent: many investigators who examined the question separately and converged, jurors polled apart, doctors who each saw the patient. In those cases each additional agreement adds information, and being the lone holdout against a genuinely independent convergence is a real reason to re-examine your position. The test is whether the agreements are separate observations or one observation repeated, and whether anyone in the group is in a position to be wrong out loud.

Built from: Logically Fallacious — Gadarene Swine Fallacy · Wikipedia — Argumentum ad populum · Wikipedia — R. D. Laing · Stanford Encyclopedia of Philosophy — Fallacies

Galileo Fallacy

Treating the fact that the experts laughed at you as proof you are right, because the experts once laughed at people who turned out to be right.

Under a video about a home protocol for treating tumours with baking soda, the top comment reads: "Every doctor here is going to tell you this is dangerous. They said the same thing about Galileo. They always attack what they can't control." Somebody replies asking for a single trial. The answer is another list of persecuted geniuses. Two hundred people upvote it, and not one of them has been given a reason to think baking soda does anything.

The tell: Every objection somehow strengthens their case. You bring evidence and it gets absorbed as more proof of the persecution, so you slowly realise there is no answer you could give that would count — which is a strange property for a true claim to have.

The move: A claim C is rejected, ridiculed, or suppressed by an established body of expertise. The arguer cites a historical figure F — Galileo, Semmelweis, the Wright brothers — who was likewise rejected and later vindicated, and infers that C is true, or that C deserves increased credence. The property C shares with F is being rejected by authority; the property projected onto C is being right. Nothing is offered about the content of C at all.

Why it fails: The reference class has been assembled after the fact and filtered by the very outcome being predicted. Vindicated outsiders are remembered because they were vindicated; the enormously larger population of rejected claims that simply stayed wrong left no monuments and no names, so the sample supporting the projection consists entirely of survivors. Carl Sagan's formulation is the whole objection in one line — they laughed at Columbus and the Wright brothers, and they also laughed at Bozo the Clown. Beyond the sampling problem, rejection is not what made Galileo right; his telescope did, and the Jesuit astronomers who first resisted his findings went on to repeat the observations and largely accept them. The pattern is also perverse as a rule of evidence: it credits a claim in proportion to how badly it is doing, so the more comprehensively it fails review, the stronger its case becomes, and no amount of counterevidence can ever count against it.

• Set the rejection completely aside. What evidence is offered for the claim itself?
• Of all the ideas mainstream experts have rejected, roughly what proportion were later vindicated? Is that base rate ever addressed?
• Is the resemblance to the historical figure evidential — same kind of observation, same kind of test — or only social, in that both were mocked?
• Is opposition being treated as a point in the claim's favour, so that more rejection would mean more credibility?

Legitimate when: The history of suppressed-but-correct science is a perfectly good reason to keep a hearing open, and pointing out that consensus has been wrong before is a legitimate argument against treating consensus as final. It is also fine as encouragement — persistence in the face of dismissal is a virtue, and Logically Fallacious notes the inspirational use explicitly. The move becomes the error only when ridicule is offered as evidence of truth rather than as a reason to go and look at the evidence.

Built from: Logically Fallacious — Galileo Fallacy · Wikipedia — Galileo affair

Gambler’s Fallacy

Believing a coin, wheel, or machine 'owes' you a different result after a bad streak, when it doesn't remember what happened last time.

Game night group chat: Jordan has lost the last five coin-flip challenges to decide who orders the pizza. 'No way I lose a sixth time,' Jordan says, offering to double the bet on the next flip. 'I'm statistically due.' But the coin has no idea Jordan lost five times in a row — it's the exact same fifty-fifty flip it always was, and it will be next time too.

The tell: The gut feeling is 'the odds have to even out soon' — but if you ask what would actually make that happen, nobody can point to any real connection between the last flip and the next one. The coin, the wheel, the dice have no memory, so there's nothing there to 'even out.'

The move: Given a sequence of trials from a fixed, memoryless process (P(A) constant and independent trial to trial), an observed run of outcomes ≠A is used to conclude P(A | history) > P(A) — i.e., the unconditioned probability of the next trial is treated as if it were conditional on the streak, when for genuinely independent trials P(A | history) = P(A) regardless of the history.

Why it fails: It substitutes an implicit 'law of averages' — the unfounded assumption that short-run frequencies must self-correct toward the long-run average — for the actual conditional probability of the next trial. It conflates two different facts: the a priori probability of a whole sequence (e.g., five heads in a row is unlikely, 1/32) with the conditional probability of the next single outcome given that the first four have already occurred (which, for an independent process, is unchanged, 1/2). Nothing in an independent, memoryless mechanism carries information from past trials into future ones, so there is no physical channel through which a 'correction' could occur.

• Are the trials generated by the same fixed, memoryless mechanism (fair die, coin, wheel) each time, with no physical link between one trial and the next?
• If you write out the actual conditional probability of the next outcome given the observed streak, does it differ numerically from the plain, unconditioned probability at all?
• Is this a finite pool sampled without replacement (cards dealt from a deck, balls drawn from an urn) where removing items really does shift the remaining odds — or is it replacement/independent trials where nothing is removed?
• Is the feeling that the outcome is 'due' tracking anything besides a wish for short-run results to match the long-run average?

Legitimate when: The reasoning is sound when the trials are actually dependent: sampling without replacement from a finite pool (cards, lottery balls already drawn), a mechanism that is physically wearing down or has been tampered with (so the streak is evidence the process itself changed), or when the streak is being used, correctly, as Bayesian evidence about an unknown parameter (e.g., a coin of unknown fairness) rather than asserted about a process already known to be fair and independent.

Built from: Wikipedia · Internet Encyclopedia of Philosophy — Fallacies

Genetic Fallacy

Deciding something is true or false just because of where it came from, instead of looking at whether it actually holds up.

Someone in the group chat shares a genuinely good budgeting tip. A friend replies, 'lol that's literally from a fortune cookie, you're taking financial advice from a cookie now?' and everyone laughs it off and moves on — nobody actually checks whether the tip itself makes sense. The advice didn't get worse because a cookie handed it out; it just stopped getting looked at.

The tell: You notice the conversation never actually touched the idea itself — no one said 'here's why that's wrong,' they just said 'look where that came from' and treated the case as closed. It feels like a joke landed, but nothing was actually checked.

The move: Claim P originated from source/circumstance S (a disreputable, embarrassing, biased, humble, or otherwise suspect origin). Therefore, P is false (or worthless) — or, in the mirror-image positive form: S is a reputable/noble/pure origin, therefore P is true (or worthy). Either way, a fact about P's genesis is substituted for an examination of P's content.

Why it fails: The truth or merit of a claim is a function of its content — what it asserts and what supports it — not of the causal or historical story behind how it came to be believed or proposed. Origin can bear on the *reliability of testimony* (a habitual liar's say-so is weaker evidence than an expert's), but even a corrupt, silly, or self-interested source can accidentally state something true, and even a noble or pure source can be wrong. The Internet Encyclopedia of Philosophy frames the defect precisely as attacking or defending a claim 'because of its origin (genesis) when such an appeal to origins is irrelevant' to the claim's actual merit. When the origin fact is treated as if it settled the question — rather than as one input to weigh alongside the content itself — the argument skips the only step that could actually verify or refute P.

• Is the origin being used as a reason to skip reading, hearing, or engaging with the claim's actual content — or merely as one factor weighed alongside that content?
• Would this exact claim, if it had come from a source everyone already trusted, be judged true or worth considering? If yes, the origin is doing work that only the content should do.
• Is the origin being cited as evidence about reliability of testimony (a legitimate use — 'the source has a track record of fabricating data') or as a verdict on truth itself (illegitimate — 'therefore it's false')?
• Has anyone actually checked the claim against evidence, or has the conversation stopped at 'well, look where this came from'?

Legitimate when: Origin is legitimate evidence — not conclusive proof — when it bears directly on the reliability of the testimony itself: a habitual fabricator's uncorroborated claim, a study funded by a party with a direct financial stake in the result, or a source demonstrated to lack the relevant competence all warrant heightened scrutiny or a demand for independent verification. The move only becomes fallacious when that scrutiny is skipped and the origin is treated as having already settled the claim's truth or falsity, rather than as a reason to check further.

Built from: Wikipedia — Genetic fallacy · Internet Encyclopedia of Philosophy — Fallacies

Gish Gallop

also known as: proof by verbosity

Firing off so many separate claims at once that nobody could possibly check them all, then treating whatever went unanswered as won.

Theo gets going at the kitchen table and in about forty seconds Ines has been told she never texts back, she was cold to his mother at Easter, she spent the tax refund without asking, she left the dishes again, she hates his friends, and she doesn't want to be here. Ines starts on the dishes — she did leave them — and Theo cuts in: "So you agree with all the rest of it." She hasn't agreed to anything. She has answered one item out of six, which is all anyone can do at that speed.

The tell: You feel behind before you have said a word. You keep picking one thing to answer while the pile in front of you grows, and it stops feeling like being argued with and starts feeling like being buried — and any single thing you do answer somehow doesn't count.

The move: In an exchange with a fixed clock, arguer A delivers a rapid series of many separate assertions — a large n, each individually contestable, none developed, none sourced. Because the cost of making an assertion is far below the cost of checking one, the respondent can address only a fraction within the format. A then treats the unaddressed remainder as unrefuted, and the unrefuted remainder as standing support for the conclusion, converting the arithmetic of the format into the appearance of a win.

Why it fails: The move establishes nothing; it manufactures a scoreboard out of a time constraint. An unanswered assertion is not a supported one — silence under a clock is evidence about the clock, and Brandolini's law names the asymmetry exactly: refuting a fabricated claim costs an order of magnitude more effort than producing it. The volume also disables the ordinary repair path in two ways. Because no item is developed, there is nothing to examine even if there were time; and because the speaker abandons each item the moment it is challenged and moves to the next, the position never sits still long enough to be pinned down, so refuting any single point buys the respondent nothing. Cumulative-case reasoning, which the gallop imitates, requires that each strand carry some weight independently; a hundred unsupported assertions do not accumulate, because a hundred multiples of nothing is still nothing.

• Count the distinct assertions made and estimate the time needed to check one of them. Could anyone have answered them all within the format?
• When a single item is refuted, does the speaker defend it, concede it, or simply move to the next one?
• Are the items presented as independently supporting the same conclusion, or is the sheer number itself the point?
• Is 'you never answered X' being used as though it established X?

Legitimate when: Presenting many pieces of evidence quickly is legitimate when each is independently checkable, sourced, and offered as part of a cumulative case the audience can verify afterwards — a written brief with citations, a review of a literature, an opening statement that flags what will be established later. Two conditions separate it from the gallop: the speaker stands behind each item when it is challenged rather than substituting the next one, and the volume is calibrated to what the format can actually absorb rather than deliberately exceeding it.

Built from: Wikipedia — Gish gallop · Wikipedia — Firehose of falsehood

Hasty Generalization

also known as: secundum quid, converse accident

Judging a whole group by the tiny handful of examples you happen to have run into.

Your friend orders from a new delivery app twice in one week, both times the food shows up cold, and she posts in the group chat: 'This app is always late and the food's always cold, never using it again, and neither should any of you.' Two orders, same neighborhood, same rainy week - and now the whole app is branded useless.

The tell: Something feels off because the number of examples is way too small for how sure the person sounds - it's the gap between 'this happened twice to me' and 'this is just how it is, always, for everyone.'

The move: A sample drawn from a population is observed to have property P in some proportion; the arguer then asserts that the population as a whole (or a future/unobserved instance) has P in that same proportion or with that same confidence, without the sample's size, selection method, or coverage actually licensing the jump.

Why it fails: The evidence base is too small, too narrow, or too self-selected to carry the weight of the claimed conclusion. The defect is a mismatch between the scope/confidence of the claim and the scope/confidence the sample can actually support - the sample may be accidentally biased (convenience sample, self-selected respondents, one time of day, one location) even when no one intends to mislead.

• How was the sample selected - deliberately/randomly, or just whoever/whatever happened to be at hand?
• Does the conclusion's scope word ('all', 'everyone', 'always', 'across the country') match what the sample actually covers, or does it jump several sizes up from a handful of cases?
• Is there a plausible unsampled subgroup or counter-case that, if included, would reverse or weaken the conclusion?
• Would the arguer accept a conclusion of the same size and confidence if the same small sample had pointed the opposite way?

Legitimate when: When the sample is genuinely random or representative and large enough for the margin of error the conclusion implies, or when the conclusion is explicitly hedged to the sample's real coverage (e.g., 'in my limited experience,' 'in this pilot batch') rather than stated as a flat universal.

Ancient: Aristotle describes an error that runs the opposite direction from the fallacy of accident: instead of wrongly applying a general rule to a special case, the reasoner takes something true of one particular case - often one held under special or qualified conditions - and asserts it as an unqualified general rule. This ancient move is the root later loosely linked to what modern logic calls hasty generalization, though the two are not identical: Aristotle's concern is dropping a needed qualification, while hasty generalization is specifically about a sample too small or skewed to support a population-wide claim. — Aristotle, Sophistical Refutations (the 'converse accident' variant of secundum quid)

Built from: Wikipedia - Faulty generalization · Internet Encyclopedia of Philosophy - Fallacies · Stanford Encyclopedia of Philosophy - Fallacies

Having Your Cake

Talking about something in a way that leaves you standing on both sides at once, so you can claim you were right however it turns out.

Half the floor asks Rhea whether the Saturday shifts are mandatory. "We value flexibility here," she says, "and we also value being there for each other when it counts." Everyone nods and nobody knows. Two weeks later Danil, who didn't come in, gets a note in his file, and Priya, who did, gets nothing at all. Rhea can point at the first half of her sentence or the second half depending on who is asking, and she never had to say a word she could be held to.

The tell: You leave the conversation unable to repeat what the person actually said. When you try to paraphrase it back you get corrected in both directions — that's not quite what I said, but I didn't say the opposite either — and you slowly realise there is nothing there to write down.

The move: Asked to take a position between P and not-P, a speaker produces discourse that endorses considerations on both sides while asserting neither, so that no commitment enters the record. A stronger variant asserts P to one audience and not-P to another. Either way the speaker retains the benefits associated with holding a position — being credited with foresight, being counted an ally — while remaining unable to be shown wrong, because whichever way the matter resolves, a reading of what was said is available that matches the outcome.

Why it fails: An argument requires a commitment that could turn out false; a position that cannot turn out false is not a position but a posture, and there is nothing there for an audience to evaluate. The defect is not vagueness in itself — honest uncertainty is often the right state to be in — but the combination of vagueness with the claiming of credit: the speaker collects what only assertion earns while carrying none of assertion's exposure. In the two-audience form the failure is sharper, since the pair of commitments is jointly unsatisfiable, which means at least one audience has been told something the speaker does not hold, and the speaker's own reasoning cannot be consistent across both rooms. In both forms the audience is deprived of the thing an audience is for, namely the chance to check a stated claim against the world, and the discussion is left with the appearance of a contribution and none of the substance.

• Write down in one sentence what the speaker is committed to. Can you actually do it?
• Set the speaker's two statements side by side. Could both be acted on at the same time?
• Whichever way this turns out, could the speaker claim afterwards to have called it? If yes, they never called it.
• Does the speaker gain something from each side that holding only one of them would cost?

Legitimate when: Genuinely withholding judgement is legitimate and often the honest answer: 'I don't know, and here is what would change my mind' is itself a stated commitment with stated conditions. So are truly conditional positions — 'if the audit shows X then we cancel, if it shows Y we proceed' — and acknowledging real merit on a side you ultimately reject. What separates these from the fallacy is that each of them forecloses claiming credit later, because each says in advance what the speaker will be found to have believed.

Built from: Logically Fallacious — Having Your Cake · Wikipedia — Weasel word

Hedging

Quietly shrinking what you claimed every time somebody pushes back, while insisting you have been saying the same thing all along.

In August, Marcus told the whole bar the Ravens were winning the Super Bowl — book it, put money on it. In November it became "I said they were the team to beat in the conference." By the divisional round it is "I said they'd have a great season, and eleven and six is a great season." Every step he insists on his record, and Deb, who took the other side back in August, can no longer find the thing she was disagreeing with.

The tell: You feel like you won the point and lost the argument. You cannot locate the moment anything changed, only that nothing you said seems to have landed, and the other person is somehow still ahead on a claim you're now sure they've abandoned.

The move: A asserts a strong claim P. Challenged, A defends a weaker claim P-minus, entailed by P and far easier to hold, and presents P-minus as what was meant all along. The impression P created in the audience is retained, while only P-minus is exposed to testing. Under further pressure the process repeats — P-minus-minus, and so on — each restatement narrower than the last and each announced as no change at all, so that the label and the credit stay constant while the content shrinks.

Why it fails: Two different claims are being run under one name, and each of the two jobs — impressing an audience and surviving scrutiny — is handed to whichever version does it better. That is an equivocation across time rather than across senses of a word, and it defeats the challenge without answering it: the objection was raised against P, the reply is offered on behalf of P-minus, and because the shrinkage is denied, the audience is left believing that P withstood examination when P was quietly dropped. It also makes the position permanently untestable, since after any outcome some version of the claim can be produced as 'what I said'. Note carefully what is not the defect: qualifying a claim is not the problem, and calibrated uncertainty is good practice. The failure lies entirely in the denial that anything moved.

• Write the original claim and the current claim side by side. Is the current one narrower?
• Did the narrowing happen before the challenge was raised, or only after?
• Is the speaker still collecting the credit, alarm, or authority that only the original, stronger version could have earned?
• Ask directly: 'so you are not claiming the strong version any more?' Do they agree, or insist nothing has changed?

Legitimate when: Revising a claim you have learned was too strong is not merely permitted but required, provided the revision is announced: 'I overstated that — what I can actually defend is the narrower version' is exactly right. Stating uncertainty up front with hedge words like 'probably' or 'in most cases' is ordinary careful speech and the linguistic sense of hedging, not the fallacy. The error is specifically the retreat presented as continuity, so the test is whether the speaker will say out loud that the earlier version has been given up.

Built from: Logically Fallacious — Hedging · Wikipedia — Motte-and-bailey fallacy

Historian’s Fallacy

also known as: present knowledge projection

Judging somebody's decision by what you found out afterwards, as though they'd had the answer sheet in front of them at the time.

The Year 5 trip to the coast went ahead and the coach came back at four with forty soaked children in it. By Monday the parents' group is unanimous: anyone could see that storm coming, the head should have cancelled. The forecast at seven that morning said forty per cent chance of showers after three, which is the forecast for roughly half the days in that month, and on every one of those days the trip would have run.

The tell: The criticism is airtight and somehow impossible to answer, and then you notice why: the person making it is holding a fact you did not have. You are being asked to defend a choice against information that arrived afterwards, and every reason you give sounds like an excuse.

The move: An agent made a decision D at time t. An evaluator standing at t+n, in possession of the outcome O, describes the evidence available at t as though it pointed to O, and then judges D against that description. The knowledge set attributed to the agent is the evaluator's own; the signals that in fact pointed elsewhere at t have been filtered out of the account by the outcome, so the record D is measured against is not the record D was made from.

Why it fails: What made the situation at t difficult was that the evidence was mixed, and knowledge of O destroys the evaluator's access to that mixture: the confirming signs become salient, the contradicting ones drop out of the account, and what remains looks like a clear signal that a competent person could not have missed. The judgement therefore tests a capacity no decision procedure possesses — foreknowledge — and so it will convict every decider who was ever unlucky and acquit every one who was ever lucky, which means it is not measuring the quality of decisions at all. Fischer's own case is Pearl Harbor: there were many indications an attack was imminent, and innumerable conflicting indications suggesting other possibilities, and retrospect preserves only the first set. The defect sits in the presupposition, before any judgement is drawn, which is why the criticism feels unanswerable — the disputed question of what the evidence showed at the time has already been settled in the framing.

• Write down what was actually known at the moment of the decision, using only sources dated before it.
• Does the criticism depend on any fact that became available only after the decision was made?
• What signals pointed the other way at the time, and does the account acknowledge any of them?
• State, before looking at the outcome, whether a careful person with that information would have been expected to choose differently.

Legitimate when: Judging a past decision harshly is legitimate when the evidence that should have changed it was available at the time and was ignored, or when the process itself was bad regardless of how things turned out — nobody checked, nobody asked, the warning was on the desk unopened. It is also entirely proper to learn from outcomes and change the rule going forward; what is not proper is charging the earlier decider with having known then what the outcome taught later.

Built from: Wikipedia — Historian's fallacy · Wikipedia — David Hackett Fischer · Wikipedia — Hindsight bias

Homunculus Fallacy

Explaining how something works by putting a tiny version of the same thing inside it, which just moves the question one step further in.

Ruth is seven and wants to know how eyes work before the light goes off. Her mum does her best: "Light comes in through the front of your eye and makes a little picture at the back, and then your brain looks at the picture." Ruth thinks about this for a second. "Looks at it with what eyes?" Her mum has no answer, and Ruth is right to be unsatisfied — the story ends exactly where it started, only smaller.

The tell: The answer sounds complete and leaves you standing precisely where you were. You notice that the question you asked fits the answer you were given just as well as it fitted the thing you asked about, which is the moment to say so out loud.

The move: A capacity C — seeing, understanding, deciding, remembering — stands in need of explanation. The account offered posits an inner subsystem or agent that exercises C upon an internal representation: the eye delivers an image, and something inside inspects the image; the brain 'reads' the signal; the mind 'looks at' the memory. The explanation thus contains an instance of the very capacity it was called upon to explain, so the original question arises again of the inner agent, and the account either regresses without end or halts by fiat.

Why it fails: An explanation of a capacity earns its keep by discharging that capacity into components that do not possess it — parts that are individually stupid, whose arrangement produces the smart behaviour. Positing an inner viewer does the opposite: asking how a person sees is answered by describing how a smaller thing inside the person sees, which is the same question with a smaller subject, so no explanatory debt has been paid. The regress is not an incidental awkwardness but the proof that nothing was reduced; each level requires another observer for the same reason the first did. Dennett's Cartesian theatre is the vivid form of the defect — a stage inside the head implies an audience inside the head — and it survives even when the dualism is stripped out, because a physical homunculus faces exactly the same demand as a ghostly one. The capacity has been presupposed rather than accounted for, which places the failure with begging the question rather than with any invalid step.

• Name the capacity being explained in one word. Does the thing doing the explaining also possess that capacity?
• Put the same question to the inner part — 'and how does that part see, or decide?' Does the account have a new answer, or the same answer?
• Can the inner posit be broken into simpler pieces, none of which has the capacity on its own?
• Is an inner audience being smuggled in by the verbs — the brain 'interprets', the mind 'looks at', the system 'understands'?

Legitimate when: Decomposing a capacity into sub-agents is standard and productive practice in cognitive science and computing, provided each sub-agent is stupider than the whole and the decomposition terminates in components that interpret nothing at all — what Dennett calls discharging the homunculi. Talk of the brain 'reading' a signal is also harmless as shorthand when a mechanical story is available underneath it. The test is only whether the regress bottoms out somewhere in parts that do not need an audience of their own.

Built from: Wikipedia — Homunculus argument · Wikipedia — Cartesian theater

Hot Hand Fallacy

also known as: hot-hand belief

Deciding somebody on a good run is now more likely to keep winning, when the run is the only thing you have to go on.

Thursday night bowling league. Marisol has thrown four strikes in a row, and when the tenth frame comes up her teammate Duane hands her the anchor spot instead of taking it himself. 'She's locked in — you don't sit somebody who's locked in.' Nobody has looked at her season average, which is ordinary, and nobody can say what changed about her arm in the last ten minutes. The four strikes are the entire case for moving her.

The tell: You notice the whole argument is made out of the last few minutes. Somebody sounds very certain about what happens next, and when you ask what changed — the grip, the lane oil, the rest, anything — there is no answer except the run itself. The streak is being treated as its own explanation.

The move: A performer or device produces outcomes from a process with some success rate p. An observed run of k consecutive successes is taken as evidence that the operative rate for the next attempt now exceeds p — that the performer is 'hot' — where the streak itself is the only evidence offered. Schematically: from Success(t-1) … Success(t-k), conclude P(Success(t) | streak) > P(Success(t)), with no independent showing that the process has any serial dependence at all.

Why it fails: Runs of that length are exactly what a stable rate produces over enough attempts; people badly underestimate how streaky ordinary randomness looks, so clustering gets read as a signal when it is the expected texture of the data. Worse, the streak is almost always noticed after it happens and then examined, which is not a fair test — the run was selected because it was a run. What the arguer has observed is clustering; what the arguer asserts is dependence, and nothing bridges the two. Note the honest complication: Miller and Sanjurjo showed that the famous 1985 study which declared the hot hand illusory had itself introduced a streak-selection bias, and on correction the basketball data do show some genuine positive dependence. So the defect named here is not believing in streaks — it is claiming an elevated rate without ever comparing the run against what a flat, unchanging rate would have thrown off.

• Is this a chance device — dice, roulette, a slot machine, a shuffled deck — with no physical channel by which one attempt could affect the next?
• Was this streak singled out after someone noticed it, or was a counting rule fixed in advance and applied across the whole record?
• Has the run been compared against how often an unchanging success rate would produce a run that long over the same number of attempts?
• Is any mechanism actually named — warm-up, fatigue, rhythm, a defensive adjustment, a change in equipment — or is the streak itself the whole of the evidence?

Legitimate when: The reasoning is sound for a skill task where a real physical or psychological channel exists (warm-up, form, fatigue, confidence) and the run has been tested against a stable-rate null with streak-selection bias corrected — Miller and Sanjurjo's reanalysis of shooting data found exactly this. It is also sound when the streak is used as evidence about an unknown ability level, rather than asserted about a mechanism already known to have no memory.

Built from: Wikipedia — Hot hand · The Fallacy Files — The Hot Hand Fallacy · Logically Fallacious — Hot Hand Fallacy

Hypnotic Bait and Switch

Getting you nodding along to a string of obvious things, then slipping in the one thing you'd have argued with.

A salesman sits at the Ortegas' kitchen table with a folder open. 'You want to keep your family comfortable, right? You'd like a smaller electric bill? You'd rather your money stayed in this house than went to the power company?' Three easy yeses in a row. 'Then let's get you signed on the twenty-five-year lease tonight — the crew's in the neighborhood Thursday.' Nothing he said about comfort or bills had anything to do with a twenty-five-year lease, but the room was already in the habit of agreeing.

The tell: You catch yourself nodding at something you have not actually thought about. Looking back, the last thing said was nothing like the things before it — it was the only one you would have pushed on, and it arrived while you were still in motion from the others. There is a small lag, and then the feeling of having been carried somewhere.

The move: A speaker asserts a run of statements S1 … Sn selected precisely because the audience will assent to them at once — truisms, shared values, matters nobody disputes — and then appends a contested claim C in the same cadence, usually joined by a connective ('and that's why', 'so of course', 'then you'll agree'). No support relation between S1 … Sn and C is ever made out. What carries C is the assent habit built by the run and transferred to the item at the end of it.

Why it fails: The opening statements are chosen for being uncontestable, and that is exactly what guarantees they carry no information about the contested matter — anything everyone already agrees to cannot discriminate between the positions actually in dispute. The persuasive work is done by momentum and rhythm: agreement becomes near-automatic, the listener stops evaluating each item separately, and the last item rides in on the pattern rather than on evidence. The connective does the smuggling, asserting a support relation that is never argued for. The reverse form runs the same machinery with a string of statements the audience will reject, so that a target claim arrives already coated in refusal.

• Strike out the opening run of agreeable statements entirely: is anything left on the table that supports the contested claim?
• Would the contested claim have been accepted if it had been stated first, on its own, with no warm-up?
• Do the agreeable statements even mention the subject matter of the contested claim, or only shared values and settled facts?
• Is the connective — 'and that's why', 'so obviously' — the only thing linking the run to the final item?

Legitimate when: The same shape is entirely legitimate when the opening statements are load-bearing: a lawyer establishing undisputed facts that the argument will actually use, a teacher recalling agreed results before the step that depends on them, a proposal restating requirements the recommendation is derived from. The test is subtraction — remove the opening items and see whether the case for the final claim genuinely weakens. If nothing is lost, they were mood, not support.

Built from: Logically Fallacious — Hypnotic Bait and Switch · Wikipedia — Bait-and-switch · Stanford Encyclopedia of Philosophy — Fallacies

Hypothesis Contrary to Fact

Stating exactly what would have happened in a version of events that never actually happened.

At Sunday dinner Renata's uncle brings up her cousin's divorce for the third time this month. 'If Tomás had taken that job in Denver, those two would still be married. That's not my opinion — that's just what would have happened.' Nobody at the table can say what Denver rent would have done to them, or what the job would have asked of his hours, or whether the trouble started long before Denver ever came up. But the whole alternate life is being reported like something he watched.

The tell: The sentence sounds like a memory rather than a guess — flat, unhedged, 'would have'. And when you try to picture that other life for yourself, a dozen other endings show up just as easily, none of which the person telling it has mentioned or seems to have considered.

The move: From the fact that some event A did not occur, the arguer asserts a definite alternative history — 'had A occurred, C would have followed' — and then reasons from C as though it were an established fact. The move supplies one branch of an unrealized past with the confidence of an observation, without a model of the causal dependence between A and C, and without accounting for the rival outcomes C′, C″ that the same altered starting point equally permits.

Why it fails: Counterfactuals are not free-floating. On the standard ordering semantics they are true when the consequent holds at the closest worlds where the antecedent holds, and which worlds count as closest depends on a similarity ordering Lewis himself called notoriously vague and context-sensitive. To assert C flatly is therefore to claim possession of a causal model detailed enough to fix an entire downstream trajectory from a single altered fact — and in historical, political, and personal cases nobody has one. Small differences in how the imagined change is filled in cascade into very different endings, which is why the honest form of most such claims is 'might have' rather than 'would have'. The damage compounds when the invented outcome is then used as a settled premise for further argument, so that an unfalsifiable speculation ends up doing the work of evidence.

• What causal account is offered for that particular outcome rather than the other outcomes the same change could equally have produced?
• Is the claim stated as 'would have' when the evidence supports at most 'might have'?
• Does the imagined outcome quietly assume everything else stayed exactly the same — and is that plausible, given that the change would have moved other things too?
• Is the invented outcome then used later in the argument as if it were something that actually happened?

Legitimate when: Counterfactual reasoning is indispensable and often rigorous. It is legitimate when the dependence is genuinely modelled or measured: a controlled experiment or well-identified quasi-experiment, a closed system with few degrees of freedom ('if the breaker had been off, the light would not have come on'), an engineering failure analysis with a mechanism traced end to end, or a claim explicitly hedged with a probability and a stated model. The defect is never the counterfactual form; it is asserting one specific unrealized outcome with a certainty no available model could deliver.

Built from: Logically Fallacious — Hypothesis Contrary to Fact · Stanford Encyclopedia of Philosophy — Counterfactuals

Identity Fallacy

Waving off what someone said because of who they are, instead of dealing with the thing they actually said.

At the tenants' association meeting Priya lays out eighteen months of water bills and shows the building was charged twice for the same month in March and again in July. Curtis cuts across her: 'You moved in two years ago. People who just got here don't get an opinion on how this building is run.' The room moves on to how long people have lived there. Nobody looks at the bills, and the double charge is still sitting on them.

The tell: You realize nothing you brought has actually been touched. The reply is about your last name, your zip code, how long you've been around — and the thing you brought is still lying there unread. The room has quietly switched from whether you're right to whether you're allowed to speak.

The move: An arguer A advances an argument G bearing on some question Q. A critic responds not to the content of G but to a feature of A's identity — group membership, background, demographic category, tenure, place of origin — and either dismisses G on that basis or holds that only members of a designated group have standing to advance or assess claims about Q. The identity feature is substituted for the evidential merits of G, and the same argument from an in-group speaker is admitted without comparable scrutiny.

Why it fails: Who advances an argument bears on whether its reasons support its claim only in one case — when the claim's warrant just is the speaker's first-person experience. Where the disputed matter rests on figures, records, studies, or plain observation that anyone could check, the speaker's category is simply not among the things that make the argument good or bad, so the response addresses something other than the argument. It is a generalized form of the circumstantial ad hominem, indexed to a class rather than to an individual's interests. Two further defects give it away: it is asymmetric, since an identical argument from an in-group speaker is accepted uninspected, which shows identity is being used as a credential rather than as evidence; and it typically denies standing rather than alleging error, so the argument never gets answered at all — it gets ruled out of order.

• Is the speaker's group membership being used to answer the argument, or to rule the speaker out of the conversation before the argument is answered?
• Does the disputed claim rest on first-hand experience only that person could have, or on evidence — numbers, records, documents — that anyone could go and check?
• Would the identical words and identical evidence from an in-group speaker draw this same objection?
• Has any specific error in the argument been named, or only a fact about who is making it?

Legitimate when: Identity is legitimately load-bearing when the claim's warrant is first-person experience — what it is like to be treated a certain way, what happened to you, what you were told when the door closed. There membership is not a credential but the evidence itself, and an outsider's contrary assertion really is worth less. It is also reasonable to weigh membership as one consideration about likely blind spots. What is never licensed is treating membership as a bar to being heard on claims that rest on publicly checkable evidence, and asking for such evidence remains in order even when experience is genuinely in play.

Built from: Logically Fallacious — Identity Fallacy · Master List of Logical Fallacies (University of Texas at El Paso) — Identity Fallacy · Stanford Encyclopedia of Philosophy — Fallacies

If-By-Whiskey

Answering a yes-or-no question by describing the same thing twice — once as a monster, once as a hero — and cheering for both.

Somebody finally asks the HOA board president where he stands on short-term rentals. 'If you mean strangers roaring in at two in the morning and treating our street like a hotel hallway, I am dead against it. But if you mean a neighbor renting out her guest room to a travel nurse so she can keep her house — I'll defend that all day.' He sits down to applause from both halves of the room. The vote next month is still a coin flip, because nobody in that room can tell you what he'll do.

The tell: Everybody claps and nobody can tell you what was said. You leave sure the person agreed with you — and so does the neighbor who wanted the exact opposite. Try to write down what they would actually do on Tuesday and the page stays blank.

The move: Asked to take a position on X, the speaker offers two conditional endorsements keyed to two descriptions of X: 'if by X you mean [X under a damning description], then I am against it; if by X you mean [X under a flattering description], then I am for it.' Both descriptions pick out the same referent; only the evaluative colouring differs. The pair is then presented as a single firm stance, with no criterion for telling which description applies to a given case.

Why it fails: The two horns are not two things but one thing described twice, so the apparent division is a partition of attitudes rather than of the world. Endorsing both leaves the stated position with no content: nothing is ruled out, no course of action follows, and no fact could settle which horn a real case falls under, because the horns are separated by tone rather than by any criterion. It is equivocation performed in the open — the term's sense shifts between the halves while the speaker claims to have been consistent throughout — and it inverts the point of drawing a distinction, which is to narrow a commitment rather than to widen it until it covers everything. The tell of success is that both camps in the room believe they were agreed with.

• Strip the loaded adjectives out of both halves: do the two halves now describe the same thing?
• Is any factual test offered for telling which half a real case falls under, or only a difference in tone?
• After the answer, can you say what the speaker would actually do — vote which way, sign which paper, spend which money?
• Would supporters and opponents both walk away believing they had just been agreed with?

Legitimate when: Drawing a distinction in exactly this shape is legitimate — indeed obligatory — when the two halves really do pick out different things and a criterion is supplied for sorting cases: 'I'll back the bond if it funds the water main; I'll oppose it if it funds the parking deck; the difference is which line item survives committee.' Genuine distinction-drawing narrows the commitment and tells you what would change the speaker's vote. If-by-whiskey widens the commitment until it covers every case and tells you nothing.

Built from: Wikipedia — If-by-whiskey · Wikipedia — List of fallacies

Illicit Contraposition

Flipping a rule around backwards — swapping the two groups and sticking a 'not' on each — when that flip doesn't hold for this kind of rule.

At the school's allergy meeting the cafeteria manager reads the rule aloud: 'Nothing on the blue tray has peanuts in it.' A dad flips it over in his notebook and reads it back to the room: 'Got it — so if it doesn't have peanuts, it's on the blue tray.' The nurse stops him there. The carrot sticks on the salad bar have no peanuts either, and neither does the apple in every lunchbox in the building. Turning the rule backwards turned a promise about one tray into a claim about the whole cafeteria.

The tell: The backwards version sounds like the same rule said another way, and it lands with the same confidence — but look at what it now covers. It has quietly started making a much bigger promise than the original. The moment you go looking for one ordinary thing that sits outside both groups at once, the flipped version comes apart.

The move: Contraposition swaps a categorical statement's subject and predicate and negates both: 'All S are P' becomes 'All non-P are non-S'. The transformation preserves truth for the A form ('All S are P') and the O form ('Some S are not P') only. The fallacy applies it to an E form — 'No S are P, therefore no non-P are non-S' — or to an I form — 'Some S are P, therefore some non-P are non-S' — where the transformation is not truth-preserving.

Why it fails: Contraposition is not primitive; it is a composite of obversion, conversion, and obversion again, and it goes through only when each step is licensed. Start from an E: 'No S are P' obverts to 'All S are non-P', which converts only per accidens to 'Some non-P are S', which obverts to 'Some non-P are not non-S' — a particular negative, far weaker than the universal negative claimed. Start from an I: 'Some S are P' obverts to 'Some S are not non-P', an O form, and O propositions do not convert at all, so the chain stops dead and nothing is derivable. The counterexamples are trivially available because complement classes are enormous: no dogs are cats, yet a chair is both a non-cat and a non-dog, which falsifies 'no non-cats are non-dogs' on the spot. The habit is imported from conditional logic, where transposition — from 'if P then Q' infer 'if not-Q then not-P' — is unrestrictedly truth-preserving, which is why the categorical restriction is so easily forgotten.

• Which of the four forms is the starting statement — 'All … are', 'No … are', 'Some … are', or 'Some … are not'? Only 'All' and 'Some … are not' contrapose.
• Have both terms been swapped AND both negated? If only the swap happened, this is illicit conversion instead, a different error.
• Can you name one ordinary thing that falls outside both classes at once — a chair, which is neither a dog nor a cat? If so, an E-form contraposition fails immediately.
• If the starting statement began with 'No' or 'Some … are', work out what actually does follow, and check whether the drawn conclusion is stronger than that.

Legitimate when: Contraposition is a legitimate immediate move for A and O forms: from 'All bats are mammals' you may infer 'All non-mammals are non-bats', and from 'Some drivers are not insured' you may infer 'Some uninsured things are not non-drivers'. The conditional cousin, transposition, holds without restriction: 'if it rained, the ground is wet' gives 'if the ground is not wet, it did not rain'. The restriction bites only on the E and I categorical forms.

Ancient: Aristotle sets out which categorical statements convert and which do not: 'if no B is A, neither can any A be B'; 'if every B is A then some A is B'; 'if some B is A, then some of the A's must be B'; but 'if some B is not A, there is no necessity that some of the A's should not be B.' Contraposition is built out of these conversions together with obversion, so the two cases later catalogued as illicit contraposition trace directly back to the conversions Aristotle here refuses — the universal affirmative that converts only in part, and the particular negative that does not convert at all. Two limits should be stated plainly: Aristotle does not discuss contraposition under that name, and the vocabulary of distribution and the label 'illicit contraposition' belong to the later, chiefly medieval and early-modern codification of his results. — Aristotle, Prior Analytics I.2 (25a14–26)

Built from: The Fallacy Files — Illicit Contraposition · Logically Fallacious — Illicit Contraposition · Wikipedia — Contraposition

Illicit Major

also known as: illicit process of the major term

Being told that one small group sits inside a big one, then announcing something about everybody else in the big one.

At the dog park, Nadia's neighbor is worked up about the new licensing rule. 'Every retriever in this park is registered with the city,' he says, 'and none of the dogs from those apartments are retrievers. So none of the apartment dogs are registered — that's the whole problem right there.' Nadia points at the beagle by the gate. His tags jingle every time he moves, he lives in unit 3B, and he has never been a retriever in his life. Being told that retrievers are all registered says nothing about who else is.

The tell: The ending is bigger than anything you were handed. You were told a fact about one small group tucked inside a larger one, and somehow the last sentence is ruling on the whole larger group. If you can picture a single member of the big group that nobody mentioned, that is the gap you were feeling.

The move: In a categorical syllogism the major term is the predicate of the conclusion. Illicit major — 'illicit process of the major term' — occurs when that term is distributed in the conclusion (as the predicate of a negative conclusion, or the subject of a universal one) while it is undistributed in the major premise. The canonical shape: 'All D are M; no C are D; therefore no C are M' — 'All dogs are mammals; no cats are dogs; therefore no cats are mammals' — where 'mammals' is the predicate of an affirmative premise and so undistributed, but the predicate of a negative conclusion and so distributed.

Why it fails: A term is distributed when the statement it sits in says something about every member of its class. 'All dogs are mammals' places the dogs somewhere inside the mammals; it says nothing about mammals in general, and in particular it does not say that dogs are the only mammals. The conclusion 'no cats are mammals' does speak about every mammal — it excludes cats from all of them. So the argument helps itself to a claim covering the whole of a class on the strength of a premise that covered only a part of it, which is why the conclusion can be false while both premises are true. The counterexample writes itself: cats are mammals that are not dogs. The general rule the move violates is one-directional — whatever is distributed in the conclusion must already be distributed in the premise it came from — and the psychological reason it slips past is that the arguer silently reads the major premise as if its converse also held.

• Find the predicate of the conclusion: that is the major term. Then locate it in the premises.
• Is the conclusion negative, or universal with that term as its subject? Either way the conclusion is saying something about every member of that class.
• In the premise where that term appears, is it the subject of a universal ('All mammals are …') or the predicate of a negative ('No … are mammals')? If neither, the premise spoke only about part of the class.
• Can you name a member of the major term's class that the premises never ruled on — a mammal that is not a dog? If you can, the conclusion has overreached.

Legitimate when: The distribution rule runs one way only: a term distributed in the conclusion must be distributed in its premise, but a term distributed in a premise may perfectly well be undistributed in the conclusion. So the same overall shape is fine once the major premise is itself universal about that term. 'No mammals are reptiles; all snakes are reptiles; therefore no snakes are mammals' distributes 'mammals' as the subject of a universal negative premise, and the negative conclusion carries without complaint.

Ancient: Aristotle takes up the first-figure premise pair in which the first term belongs to all of the middle while the middle belongs to none of the last, and rules that 'there will be no syllogism in respect of the extremes' — nothing follows about the outer terms. He proves it as he proves all such rejections, by supplying concrete term-triples: with animal, man, horse the premises hold (every man is an animal, no horse is a man) and every horse is nonetheless an animal; with animal, man, stone the same premises hold and no stone is an animal. Since the premises are compatible with opposite verdicts about the outer terms, no conclusion follows from them. That rejected pair is exactly the shape later named illicit major. Be precise about the limits: Aristotle rejects the pattern and shows why, but the vocabulary of distribution and the name 'illicit process of the major term' are later scholastic codifications of what his counterexamples establish. — Aristotle, Prior Analytics I.4 (26a2–9)

Built from: Wikipedia — Illicit major · The Fallacy Files — Illicit Major · Wikipedia — List of fallacies

Illicit Minor

also known as: illicit process of the minor term

Being told a small group sits inside a big one, then announcing the whole big one is like the small one.

At the community garden's spring meeting, Ruth gives the plot report. 'Everybody on the Saturday work crew paid the water fee, and everybody on the Saturday crew has a corner plot. So all the corner plots are paid up — we can stop chasing them.' Except corner plots were handed out by lottery back in February, long before anyone signed up for Saturdays, and three of them belong to people Ruth has never met. What she knew was that the Saturday crew have corner plots. She did not know that having a corner plot means Saturday crew.

The tell: Two solid-sounding sentences about the same small group, and then an ending that suddenly covers a much bigger group — the one that only ever got mentioned in passing. The safe version of that ending would have started with the word 'some', and nobody said 'some'.

The move: In a categorical syllogism the minor term is the subject of the conclusion. Illicit minor — 'illicit process of the minor term' — occurs when that term is distributed in the conclusion, the conclusion being universal about it, while it is undistributed in the minor premise, typically because it sits there as the predicate of an affirmative statement. The canonical shape: 'All M are P; all M are S; therefore all S are P' — 'All cats are felines; all cats are mammals; therefore all mammals are felines' — where 'mammals' is undistributed as the predicate of an affirmative premise but distributed as the subject of a universal conclusion.

Why it fails: The minor premise 'All cats are mammals' locates the cats inside the mammals and says nothing whatever about the mammals not caught up in that placement. The conclusion 'All mammals are felines' makes a claim about each and every mammal. Since nothing entitled anyone to think the cats exhaust the mammals, the dog is standing right there as a counterexample with both premises still true. Formally this is why the third figure with two universal affirmative premises yields only a particular conclusion — 'some mammals are felines' is what the premises actually buy — and pushing that to a universal is exactly the illicit process. The move usually rides on the same silent error as its twin: reading an affirmative premise as though its converse also held.

• Find the subject of the conclusion: that is the minor term.
• Is the conclusion universal about it — 'all X are …', 'no X is …'? Then it is speaking about every member of X.
• In the premise where that term appears, is it the subject of a universal, or merely the predicate of an affirmative ('all cats are mammals')? A predicate of an affirmative covers only part of its class.
• Weaken the conclusion from 'all' to 'some' and reread the argument. If the weakened version is fine and the original is not, the offence is the extra reach.

Legitimate when: The same premises support the weaker particular conclusion, and that one is legitimate: 'All cats are felines; all cats are mammals; therefore some mammals are felines' is the valid third-figure mood Darapti, on the traditional reading that treats the shared subject class as non-empty. More generally, the rule bans only distributing in the conclusion what a premise left undistributed — whenever the minor premise is itself universal about the minor term, a universal conclusion is licensed.

Ancient: Setting out the third figure — where both premises share the same subject term — Aristotle states what follows when both are universal affirmatives: 'whenever both P and R belong to S, it follows that P will necessarily belong to some R.' The conclusion he licenses is particular: some R, not all R. That boundary is precisely what illicit minor oversteps, since the fallacy takes the same two premises and reads off a universal conclusion about the minor term. State the limit honestly: Aristotle marks the boundary by declaring what does follow and declaring it as a particular; he does not name a fallacy here, and the vocabulary of distribution and the label 'illicit process of the minor term' are later scholastic codifications. — Aristotle, Prior Analytics I.6 (28a17–26)

Built from: Wikipedia — Illicit minor · The Fallacy Files — Illicit Minor · Wikipedia — List of fallacies

Illicit Substitution of Identicals

Deciding two things must be different people, because somebody didn't recognize one of them.

The family group chat is still trying to work out who left five hundred dollars in the collection envelope at Grandma Ines's funeral. Marcus writes: 'It wasn't Uncle Ray. Grandma knew Ray her whole life, and she told me once she had no idea who kept slipping money into that envelope.' Everyone accepts it and the chat moves on to the catering bill. But Grandma not knowing who was leaving the money doesn't make Ray a different man than the one leaving it — she could know Ray perfectly well and still not know Ray was the one doing it.

The tell: The whole argument turns on what somebody did or didn't recognize, but the ending is about the world — about who a person actually is. That swap is the itch. Failing to place a face has never once changed whose face it was.

The move: Leibniz's law licenses substituting co-referring terms salva veritate: from a = b and Φ(a), infer Φ(b). The fallacy applies that rule inside a referentially opaque — intensional — context: 'believes that', 'knows that', 'hopes', 'is looking for', 'is surprised that', 'necessarily', or quotation. Schematically, a = b; S believes that Φ(a); therefore S believes that Φ(b). It more often appears in the contrapositive dress the classic puzzles wear: S knows a; S does not know b; therefore a ≠ b.

Why it fails: Inside an attitude report the sentence is not simply about the object; it is about how the object is presented to the subject. Lois Lane believes Superman can fly and does not believe Clark Kent can, and both reports are true, which shows that something beyond the reference of the names is fixing the truth conditions — the names' sense, on a Fregean account, or the mode of presentation under which the believer grasps the individual, on a neo-Russellian one. Since the identity a = b can hold in the world without holding in the subject's head, substituting one term for the other can turn a true report into a false one, and truth-preservation is exactly what the rule was supposed to guarantee. In the contrapositive form the defect is starker still: a person's failure to recognize an identity is a fact about that person, and no fact about a person's knowledge can make an identity in the world false.

• Is the swapped term sitting inside a verb of attitude — believes, knows, hopes, fears, is looking for, is surprised that — or inside quotation marks, or under a 'necessarily'?
• Is the identity known to the person whose state of mind is being reported, or only to the arguer making the argument?
• Rewrite the claim with the attitude verb stripped out: does the substitution still look harmless? If the trouble appears only when the verb is restored, the context is what is doing it.
• Does the argument end by concluding that two things are different because somebody failed to recognize one of them? That is this move's signature.

Legitimate when: Substitution is perfectly good in extensional contexts, where a sentence's truth turns only on which objects the terms pick out: from 'Clark Kent is Superman' and 'Clark Kent lives in Metropolis' you may infer 'Superman lives in Metropolis', and no puzzle arises. The restriction is not on identity but on position — the swap must fall outside the scope of an attitude verb, a modal operator, or quotation marks. Even inside an attitude report, substitution is fine when the report is explicitly de re ('of Clark Kent, she believes that he can fly').

Ancient: Diogenes Laertius credits Eubulides of Miletus with a set of puzzles that includes the Veiled Man (enkekalymmenos) and the Electra. In the Electra: Electra knows her brother Orestes; the man approaching her is Orestes; but she does not know the man approaching — so does she know him or not? The puzzle isolates the exact position where swapping co-referring names inside 'knows' stops preserving truth. The attribution is not clean and should not be reported as if it were: Diogenes elsewhere, at II.111, credits the Veiled and the Horned arguments to Diodorus Cronus, and the fully elaborated wordings descend through later sources. The puzzles are securely ancient; their author is disputed within the ancient testimony itself. — Diogenes Laertius, Lives of Eminent Philosophers II.108 (the puzzles of Eubulides of Miletus)

Built from: Wikipedia — Masked-man fallacy · Wikipedia — Eubulides · Stanford Encyclopedia of Philosophy — Propositional Attitude Reports

Imposter Fallacy

Insisting the embarrassing people in your group must be enemy plants, with nothing at all to show for it.

After Saturday's match, three people in the supporters' section threw bottles at the away team's bus, and it led the eleven o'clock news. By Sunday morning an admin has posted in the club forum: 'Those weren't ours. Nobody in this section behaves like that — they were City fans in our colors, sent in to make us look bad, plain and simple.' He has no names, no footage, no source. Two of the three have had season tickets in that block for six years, and one of them is in the group photo pinned to the top of the forum.

The tell: Notice how fast the story arrived and how little it is carrying. The 'they were plants' explanation shows up the moment the news is bad and never when the news is good — and no matter what you put in front of them, the bad ones always turn out to be somebody else's.

The move: A group G is embarrassed by the conduct of members X, Y, Z. A defender of G asserts, without evidence gathered independently of the embarrassment, that X, Y, Z were never really members of G but infiltrators from a rival group planted to discredit it, and concludes that their conduct tells us nothing about G. The infiltration claim is introduced after the fact and is supported by nothing beyond the need to explain the fact away.

Why it fails: The infiltration hypothesis is adopted because it is convenient rather than because anything supports it, which is the structure of an ad hoc rescue: the generalization about G survives only because a rescuing claim was minted on demand. As deployed it is also effectively unfalsifiable, since any future counter-instance can be reassigned to the enemy by the same manoeuvre, and a claim no evidence could disturb is not doing evidential work. It fails a consistency test as well — the same authenticity standard is never applied to members whose conduct flatters the group, which shows the standard is tracking convenience rather than membership. And it inverts the burden of proof, treating infiltration as the default reading of any embarrassing act and demanding that opponents prove genuine membership.

• What evidence supports the claim that these people were plants, other than the fact that their conduct was embarrassing?
• Was the authenticity of these particular people ever questioned before they became inconvenient?
• Apply the same authenticity test to a flattering example — a member whose conduct made the group look good. Would it disqualify them too?
• If the infiltration claim turned out to be false, is there anything the arguer would then accept as showing the conduct came from inside the group?

Legitimate when: Infiltration is a real phenomenon and is sometimes documented: undercover officers, agents provocateurs, brigading accounts, astroturfed commenters, planted hecklers. The claim is legitimate exactly when it rests on evidence gathered independently of the embarrassment — identified accounts, court records, payment trails, admissions, forensic traces — and the practical test is whether that evidence would have been persuasive before the awkward incident occurred. Asserting infiltration first and looking for evidence afterward, if at all, is what makes it the fallacy.

Built from: Logically Fallacious — Imposter Fallacy · Wikipedia — No true Scotsman · Stanford Encyclopedia of Philosophy — Fallacies

Incomplete Comparison

Saying something is better, faster, or cheaper without ever saying better, faster, or cheaper than what.

Wandering the mattress showroom on a Saturday, Ellis reads the tag hanging off the one his wife likes: 'Sleep up to 43% cooler.' He asks the salesman cooler than what — the model beside it? A regular mattress? Sleeping outside? The salesman says, cheerfully, 'Cooler.' The tag doesn't say. The brochure doesn't say. And the number can't be wrong, because there's nothing on the other side of it for it to be wrong about.

The tell: You believe the claim, and then you notice you can't repeat it to anybody. Try saying it back out loud — 'it's cooler than…' — and the sentence just runs out. Whatever you were picturing at the other end of it, you put there yourself.

The move: A comparative expression — 'better', 'faster', 'more', 'up to 40% less' — is asserted with one relatum suppressed: 'X is better' without a statement of what X is better than, on what measure, over what range, or under what conditions. The audience supplies a flattering default comparison class; the speaker remains answerable only for the weakest comparison that would make the sentence true.

Why it fails: A comparative expresses a two-place relation, and with one place unfilled it does not express a determinate claim at all — there is nothing to test, nothing to check, and nothing that could count as refuting it. Because the audience fills the gap while the speaker does not, the sentence's persuasive content and its defensible content come apart: a shopper reads '40% less fat' against a reasonable alternative, while the fine print secures it only against the fattiest bar on the shelf. That gap between what is heard and what is said is why this belongs with the ambiguity family rather than with faulty comparison — the defect is not that two things were badly compared but that only one thing was ever named.

• Compared with what? Is the second item of the comparison stated anywhere at all, including the fine print?
• On what measure, over what period, and across what sample — and is that stated too?
• Would the sentence still be true against the weakest possible comparison — 'more than nothing', 'better than the worst one on the market'? If so, that may be all it actually claims.
• Can you describe a result that would make the claim false? If no such result exists, no claim has been made yet.

Legitimate when: A comparative is complete and legitimate when the second term is supplied explicitly, or is genuinely fixed by context: 'do better in school' plainly means better than your own last term, and 'the cheaper of these two' is complete standing in front of the shelf. Honest advertising names the comparison — 'costs less than the leading brand at the same weight' — and honest research states the control. The test is whether a reader could in principle go and check.

Built from: Wikipedia — Incomplete comparison · Logically Fallacious — Incomplete Comparison · Wikipedia — List of fallacies

Inconsistency

Holding on to two things that can't both be true, and leaning on whichever one is handy at the moment.

The dent in the passenger door comes up at breakfast. Theo says he was at Amara's all night and never touched the car. Then he says the dent was already there when he got in it. Then he says a shopping cart got it in the Kroger lot while he was inside. His mother sets down her coffee. Any one of those, on its own, would have gotten him off the hook. All three together tell her he was in the car — and that at least two of the stories were made up on the spot.

The tell: You can't get hold of it, because it keeps moving. Knock down one answer and the next one is already waiting, and the next one contradicts the last. The feeling is of being handed three keys and told they all open the same door.

The move: An arguer asserts, or is committed to, a set of claims {P1 … Pn} that cannot all be true together, and presses each as though the others were not on the table. Two shapes recur. Internal contradiction: a single statement conflicts with itself — 'nobody goes there anymore, it's too crowded.' Inconsistent defence: several individually respectable answers to one charge are stacked, though accepting any one of them falsifies the rest.

Why it fails: An inconsistent set guarantees that at least one of its members is false, so the arguer's own commitments establish that some of what is being asserted is untrue — without indicating which. That destroys the set's usefulness as support: a contradiction entails everything, so it discriminates nothing, and an audience cannot tell which claim is being relied upon at any given moment. Rhetorically that indeterminacy is the point, since each objection is met with whichever branch is not currently under attack, and the arguer never has to choose. The demand being made here is not that a person's views be attractive, only that they be jointly satisfiable — a set of claims that could not all be true cannot jointly support anything.

• Write each commitment on its own line: is there a single way the world could be under which all of them hold?
• If one of them is true, does it make another one false? Ask which one the arguer is prepared to drop.
• Is a qualification of time, scope, or jurisdiction available that would reconcile them — and if so, why was none offered when the claims were made?
• Is the same principle being applied to one case and its opposite applied to another, with only the identity of the parties differing?

Legitimate when: Apparent inconsistency dissolves legitimately when the claims are indexed differently — different times ('I opposed it in March; I support the amended version now'), different scopes, different senses of a shared word — or when they are offered explicitly as alternatives, in the alternative: 'we deny the debt; and should the court find otherwise, it is time-barred.' That is ordinary honest pleading precisely because the alternatives are labelled as alternatives rather than all asserted flatly, and because the arguer can say which one they believe.

Ancient: Socrates draws out two things Euthyphro holds: that the pious is what the gods love, and that the gods quarrel and love opposite things. Taken together these entail that one and the same act is both pious and impious, which Euthyphro will not accept. Socrates offers no rival definition; the joint unsatisfiability of Euthyphro's own commitments is itself the result, and the method — eliciting commitments and then exhibiting that they cannot all stand together — is the Socratic elenchus at work. — Plato, Euthyphro (7e–8b)

Built from: Logically Fallacious — Inconsistency · Internet Encyclopedia of Philosophy — Fallacies · Wikipedia — Kettle logic

Inflation of Conflict

Treating a small argument among the people who study something as proof that none of them know anything.

At the block party, Dev's brother-in-law waves off what his doctor told him about his blood pressure. 'One year eggs are killing you, next year they're a superfood. Those nutrition people can't agree on anything, so it's all guesswork — I'll eat what I want.' The salt advice he is ignoring has been steady for forty years. The egg argument he is citing is a narrow one about cholesterol in people with a particular condition. One small disagreement has been promoted into a blanket permission slip.

The tell: The argument being pointed at is real, which is what makes it hard to push back on — but somehow it has grown to cover everything, including the part nobody was arguing about. And the growth always ends in the same place: clearing the way to do whatever the person already wanted to do.

The move: Experts in a field F are observed to disagree about some point p within F. From that observation the arguer concludes that F yields no knowledge at all, or that F's settled results may be disregarded, or that a position lying outside the range of expert disagreement is thereby legitimated. Two shapes: the disagreement is genuine but peripheral and is treated as though it were central, or the disagreement is manufactured by giving an outlier equal billing with a settled consensus.

Why it fails: Disagreement at the frontier is what an active discipline looks like, and it presupposes rather than undermines the large body of agreed results that make the disputed question askable in the first place. The inference generalizes from one narrow live controversy to a verdict on the whole field — an enormous leap from a very small base — and it ignores both the proportion and the location of the dispute: how many working specialists hold each view, and whether the dispute touches the claim actually at issue or something well downstream of it. The move also proves far too much: applied consistently it would dissolve medicine, aviation, and structural engineering, which the arguer relies on without hesitation. And the further step from 'the experts differ' to 'so my position is as good as theirs' adds a second failure, since a range of expert estimates confers nothing on a position outside that range.

• Is the disagreement about the specific claim under discussion, or about a detail, an estimate, or a mechanism downstream of it?
• What proportion of working specialists holds each side — is this a live split in the field, or one dissenter given equal airtime?
• Does the position being defended fall inside the range of expert disagreement, or outside it entirely?
• Applied to aviation or emergency medicine, would this same standard let you disregard those fields too?

Legitimate when: Pointing to expert disagreement is a good argument when the disagreement is genuine, widespread, and about the core claim, and when the parties are bona fide specialists. Then a non-specialist has real reason to suspend judgement — a much weaker and much more honest conclusion than 'nothing is known here.' Reporting a genuine range of expert estimates as a range, rather than collapsing it to a point, is also correct and not this move.

Ancient: The Pyrrhonist mode from disagreement observes that on a given matter there is undecided conflict, both in ordinary life and among the philosophers, and that the competing views appear of equal weight. From this Sextus draws suspension of judgement — epoche — and continued investigation. He is emphatic that this is not the claim that the matter is unknowable, which he classes as negative dogmatism and rejects. That is exactly the boundary inflation of conflict crosses: it takes the same starting observation and hardens the Pyrrhonist's suspension into the far stronger verdict that a whole field yields nothing. The ancient text is the ancestor of the move, and the honest version of it, not an instance of the fallacy. — Sextus Empiricus, Outlines of Pyrrhonism I (the mode from disagreement, diaphonia)

Built from: Logically Fallacious — Inflation of Conflict · Wikipedia — List of fallacies · Stanford Encyclopedia of Philosophy — Ancient Skepticism

Insignificant Cause

Blaming a real but tiny contributor for the whole outcome, as if fixing it would fix everything.

Money is tight again and Cass's dad has found the culprit at the kitchen table. 'Four streaming services. Forty dollars a month. That's your problem right there — cancel those and you're fine.' The forty dollars is real; nobody's disputing it. But rent went up three hundred in March, car insurance ninety, and daycare two-forty. Cancel every subscription in the house and the hole is still four hundred and fifty deep. The conversation, though, is over.

The tell: The thing being blamed is real, which is exactly what makes it hard to push back on — and it is also strangely small next to the mess it is supposed to explain. Then notice what isn't in the sentence. The bigger numbers usually belong to someone who would rather not be discussing them.

The move: An effect E has several contributing causes C1 … Cn of widely differing magnitude. The arguer identifies one genuine but minor contributor Ck and presents it as the cause of E — 'this is why it happened' — usually drawing the practical conclusion that removing Ck removes E. The causal contribution is real; what is misstated is its share.

Why it fails: Causal claims about a particular outcome are quantitative, not merely existential. That Ck is among the causes licenses 'Ck made some difference'; it does not license 'Ck made the difference'. The counterfactual test settles it directly: remove Ck, hold everything else fixed, and ask how much of E remains — if nearly all of it does, the factor cannot carry the explanatory weight placed on it. The move usually rides on a bias toward matching large effects to memorable causes, and it is expensive in practice, because acting on the trivial factor consumes exactly the attention and budget the dominant ones needed. It also frequently serves an interest: the causes left unmentioned tend to be the ones inconvenient to the speaker. Note that the label misleads slightly. The point is not that small causes never suffice — a single spark in a room full of fumes does — but that this factor's measured share is inadequate to the effect being attributed to it.

• Roughly what share of the effect can the named factor account for, set beside the other contributors?
• If the named factor were removed entirely and nothing else changed, how much of the effect would still be there?
• Has the factor's magnitude been measured, or has only its existence been noticed?
• Which larger contributors go unmentioned, and would naming them be inconvenient for the person making the claim?

Legitimate when: A small cause is correctly named as the cause when it is genuinely the difference-maker — the last straw on an otherwise loaded system, the spark in a room already full of fumes, the one loose bolt — or when the claim is stated proportionally: 'this accounts for roughly a tenth of the gap.' The test is measurement rather than size. A factor can be tiny in absolute terms and still be the one that had to be there, and saying so is not this fallacy.

Built from: Logically Fallacious — Insignificant Cause · Wikipedia — Fallacy of the single cause

Jumping to Conclusions

Settling on one story about what happened before anything has actually been checked.

Wren sends a long message to her friend Iyla on Tuesday morning and hears nothing back. By that evening she has the whole thing worked out: Iyla is angry about the barbecue, she has been talking to their other friends about it, and this is her easing Wren out. On Thursday Iyla writes back — she cracked her phone screen Monday night and the replacement took three days. Two of those days Wren spent answering Iyla's side of an argument that never happened.

The tell: Notice how fast the certainty arrived and how little it is standing on. The story explains everything, has no loose ends, and was assembled in about four seconds — and if you try to write down what you actually saw, as opposed to what you decided it meant, the list is embarrassingly short.

The move: An arguer moves from a small quantity of evidence — often a single observation, and often an ambiguous one — to a confident, specific claim that the evidence would license only after further checking. The characteristic step is that an interpretation is filed as an observation: what was actually seen and what was read into it are conflated, and the reading is then reasoned from as though it were data. Recurring species: mind-reading (attributing a thought or motive), fortune-telling (a fixed prediction of a bad outcome), and premature diagnosis.

Why it fails: Closure arrives before the evidence discriminates. Several explanations of the observation remain live, what is in hand does not favour the chosen one over its rivals, and no attempt is made to gather the cheap further evidence that would separate them — the one question, the one look. Because the chosen reading is then stored as a fact rather than as a guess, subsequent evidence gets read in its light, and the first interpretation becomes the frame everything else is fitted into, which makes the error self-sealing. The defect is one of proportion between evidence and confidence rather than of the argument's shape: the identical step, taken tentatively and revisited when more arrives, would be ordinary good practice.

• Separate what was actually observed from what was read into it — write the two on separate lines and compare their lengths.
• Name two other explanations of the same observation. Does the evidence in hand rule either of them out?
• What single cheap check — one question, one phone call, one look — would settle it, and why has it not been done?
• Is the reading now being carried forward as a fact, so that new information is being interpreted through it rather than tested against it?

Legitimate when: Quick judgement on thin evidence is right when the cost of delay exceeds the cost of being wrong — a clinician treating a suspected stroke, a driver braking for movement at the roadside — provided the judgement is held provisionally and revised as more comes in. It is also fine when the single observation genuinely is diagnostic and rules out the alternatives by itself. What makes it the fallacy is confidence unmatched to evidence, plus the refusal of a check that was available and cheap.

Built from: Wikipedia — Jumping to conclusions · Internet Encyclopedia of Philosophy — Fallacies · Logically Fallacious — Jumping to Conclusions · Wikipedia — List of fallacies

Just Because Fallacy

Answering "why?" with "because I said so" — treating your own say-so as if it were the reason.

At Tuesday marching band practice, Coach Delgado tells the drumline they're switching to a slower tempo for the whole show. Marcus, the section leader, asks what changed — the judges' sheet, the pit, something with the horns? Delgado says, "Because I said so. Next." Marcus tries once more: "I'm not arguing, I just need to tell my section something." Delgado repeats it — "Because I said so" — and the section spends the rest of the week guessing at a reason nobody ever gave them.

The tell: You asked a real question and got a full stop instead of an answer, and you feel oddly scolded for having asked. Notice that you know exactly as much now as you did before you opened your mouth — the words that came back had no content in them, only volume. There is often a little flush of embarrassment, as though wanting a reason were itself rude.

The move: An arguer A asserts a claim C. Challenged to supply grounds, A offers no independent statement bearing on C's truth, and instead returns a formula — "just because", "because I said so", "trust me", "you'll see" — whose only content is that A has asserted C. The support slot in the argument is filled by the fact of the assertion itself, or by A's standing as the one asserting it, so what is offered as a reason for C is a restatement that C has been claimed.

Why it fails: A reason has to be something a listener could evaluate apart from the claim it supports; "just because" supplies nothing of that kind. What it substitutes is either a bare repetition of the claim under emphasis, which begs the question, or an appeal to the speaker's own say-so, which is the classical ipse dixit — the Pythagoreans' autos epha, 'he himself said it', reported by Cicero as the pattern of adopting a view because the master pronounced it. Either way no evidence has entered the exchange, so the listener's epistemic position is exactly what it was before the challenge. The move is also perfectly reversible: the same formula would support not-C with equal ease, which is the surest sign that it is doing no discriminating work. And because it forecloses the challenge rather than answering it, it converts an argument into a demand for compliance, which may sometimes be an appropriate exercise of authority but is never an argument.

• When asked for grounds, does the arguer supply any statement that could be checked independently of the claim itself?
• Is the entire support offered some form of 'because I said so', 'trust me', 'just because', or a restatement of the claim with added emphasis?
• Would the identical formula work just as well to support the opposite claim? If yes, it is not functioning as a reason.
• Is the speaker exercising legitimate authority to close a decision (a parent, a chair calling a vote, a commander) rather than purporting to have argued for it — and is the difference being acknowledged?

Legitimate when: Closing a matter by authority is legitimate when the speaker actually holds the relevant decision right and is issuing a directive rather than an argument — a parent setting a bedtime, a referee ruling, a surgeon in theatre. It is also fine to defer a reason honestly, saying 'I have grounds I can't share right now', because that leaves the claim open to later challenge instead of pretending it has been supported.

Ancient: Cicero reports that the followers of Pythagoras settled disputes by saying autos epha — 'he himself said it' — treating the master's bare pronouncement as sufficient warrant, with no argument required. Cicero renders this as ipse dixit and objects that authority ought not to prevail in matters where reasons are to be sought, which is the earliest clean statement of the defect. — Cicero, De Natura Deorum I.10

Built from: Logically Fallacious — Just Because Fallacy · Wikipedia — Ipse dixit · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies

Just In Case Fallacy

Treating "it could happen" as if it settled the matter, without ever asking how likely it is or what the safeguard costs.

Priya and Wes are at the electronics counter buying a $180 blender, and the clerk offers a $70 four-year protection plan. Wes says no thanks. The clerk leans in: "Sir, these motors can burn out. Imagine you're mid-smoothie on a Saturday morning and it just dies." Wes asks how often that actually happens, and the clerk says, "It only takes once." On the drive home Priya points out that they've just been talked toward spending forty percent of the blender's price against a number nobody would name.

The tell: You notice the other person keeps painting the disaster in more detail instead of telling you how likely it is. Every time you ask for a number, you get another version of the story. Your stomach is being argued with while your head is left waiting, and if you agree you'll walk away still not knowing the one thing that would have made the decision easy.

The move: An outcome H would be very bad if it occurred. H cannot be ruled out — its probability is left unstated, or is treated as irrelevant once the badness of H is on the table. A costly precaution M is then said to be required, on the ground that M would avert H. Neither P(H) nor the costs of M, nor the harms M itself introduces, enter the calculation; the argument runs from the mere possibility and the vividness of H directly to the necessity of M.

Why it fails: Precautionary reasoning is a comparison, not a reflex: the expected cost of accepting the risk has to be weighed against the full cost of the precaution, and both terms need a probability attached. Dropping the probability term makes the argument insensitive to whether the feared outcome is likely or one-in-a-billion, which means it licenses the same conclusion in cases that are not remotely alike. It also has no stopping rule — indefinitely many bad outcomes cannot be ruled out at any moment, and the same standard would mandate precautions against all of them at once, including precautions that contradict each other. Finally, the precaution itself has costs and often its own risks, so a measure adopted 'just in case' can raise total expected harm rather than lower it. The argumentative work is being done by fear of the imagined scenario rather than by any estimate of how the world actually is.

• Is any estimate of how likely the feared outcome is offered, or only a demonstration that it cannot be ruled out?
• Are the costs of the proposed precaution weighed at all — including harms and risks the precaution itself introduces?
• Would the same 'we can't rule it out' standard also require precautions against other unlikely harms, including ones incompatible with this one?
• If the feared outcome were made ten times less likely, would the argument as stated change in any way? If not, probability is not in it.

Legitimate when: Genuine precautionary reasoning is sound when the stakes, the likelihood, and the price of the precaution are all on the table and the trade favours acting: cheap insurance against a catastrophic and irreversible loss is rational even at low probability, which is why we wear seatbelts and back up our files. The formal precautionary principle is defensible in exactly this shape — where harm is severe, plausible and irreversible, and the safeguard is proportionate — and it collapses into the fallacious version only when the probability and cost terms are dropped.

Built from: Logically Fallacious — Just In Case Fallacy · Wikipedia — Precautionary principle · Wikipedia — Appeal to fear · Stanford Encyclopedia of Philosophy — Risk

Kettle Logic

Piling up excuses that can't all be true at once, and hoping the size of the pile counts as strength.

The group project is due at midnight and Tara's slides never landed in the shared folder. In the chat she writes: "I uploaded them yesterday. Also the folder link you sent was broken. Also nobody ever told me I had the slides." Devon reads it twice and replies, "Any one of those would be fine. All three can't be." The chat goes quiet, because the pile that was meant to look overwhelming has just told everyone that none of it is the real story.

The tell: The more excuses arrive, the less you believe any of them — and you can't immediately say why, because each one on its own sounded fine. Try holding two of them in your head at the same time and you'll feel them shove each other. There's also a distinct sense of being buried rather than answered, as though volume were being offered in place of a straight account.

The move: Facing a charge or objection O, an arguer A offers defences D1, D2, ... Dn, stacked as though their support were cumulative. But the set is not jointly satisfiable: the truth of D1 requires the falsity of D2, and so on. A does not choose among them or arrange them as ranked alternatives; A asserts them together, so that the appearance of strength comes from the number of defences rather than from any one of them surviving scrutiny.

Why it fails: Independent reasons for one claim really do accumulate, and an arguer is entitled to that; kettle logic borrows the look of accumulation while destroying its basis. If the defences contradict one another, then at most one can be true, so the aggregate does not raise the credibility of the position above the credibility of its best single member — and it lowers it, because an arguer who sincerely believed any one of them would not need the others. The inconsistency is itself evidence about the arguer: it shows the defences were selected for their power to deflect rather than because they are believed, which is exactly what an accuser is trying to establish. There is a further structural point. From a contradictory set anything whatever follows, so a listener who accepts the whole bundle has been handed a licence to derive both the arguer's innocence and his guilt.

• Take the defences two at a time: can both be true at once, or does each require the other to be false?
• Are they offered as ranked alternatives ('primarily X; and even if not X, then Y') or asserted flatly side by side as though all held?
• Would the arguer stake the whole case on any single one of them if the others were struck out?
• Is the persuasive force coming from the number of defences rather than from the strength of the best one?

Legitimate when: Genuine cumulative cases are fine — several compatible, independently supported reasons for the same conclusion really do add up. Explicit pleading in the alternative is also legitimate and is standard in law: 'my client was not there; in the alternative, if he was there, he acted in self-defence' is not an assertion of both, but a conditional structure that names which claim is primary and which is a fallback should the first fail. What marks the fallacy is asserting the incompatible defences flatly, together, as if each added weight to the others.

Built from: Wikipedia — Kettle logic · Logically Fallacious — Kettle Logic · Wikipedia — Alternative pleading · Stanford Encyclopedia of Philosophy — Aristotle on Non-contradiction

Least Plausible Hypothesis

Reaching past the two or three ordinary explanations for the far-fetched one, just because it would also fit what you saw.

The wifi drops out on a Thursday night and Callum announces to his flatmates that it's a government signal test — he read that they run them regionally, and it would explain the outage perfectly. Bea points out that the storm knocked out power on the next street, that their router is nine years old, and that the bill is three weeks overdue. Callum says none of that proves it isn't the signal test. It doesn't; it just means he has walked past three explanations that need nothing extra to be believed in order to reach one that needs a secret programme nobody has evidence of.

The tell: The story you're being handed is more interesting than the situation deserves, and it arrives fully formed with extra machinery attached. When you offer the boring explanation, the answer isn't evidence against it — it's "well, you can't prove it wasn't." You'll notice you're being asked to accept several new things at once, none of which came with any support of its own.

The move: Evidence E is observed. Several hypotheses H1 ... Hn would each account for E. The arguer settles on the member of that set with the lowest prior plausibility and the largest number of unevidenced auxiliary assumptions, on the ground that it too would explain E, and treats E as support for it. The comparative step that gives explanatory reasoning its force — ranking the rivals against one another — is either skipped or run backwards.

Why it fails: Inference to the best explanation licenses a hypothesis only by comparison: E supports H to the extent that H accounts for E better than its rivals do. Since a great many stories can be told that would produce the same observation, the bare fact that H would explain E is almost no evidence at all — it is a condition of entry, not a victory. Choosing the least plausible member of the set therefore takes the one property that all candidates share and treats it as though it discriminated among them. Worse, the far-fetched candidate typically buys its fit by adding assumptions that are themselves unsupported, and each such addition multiplies into a lower prior, so the very manoeuvres that make it cover E make it less likely to be true. Parsimony is not an aesthetic preference here; it tracks the arithmetic of conjoined improbabilities. What usually drives the selection is desire, dread, or loyalty to a prior commitment, none of which is evidence.

• List the rival explanations of the same observation: has the arguer named any of them, or only the one favoured?
• Count the assumptions each rival needs that have no independent support of their own — how does the favoured one compare?
• Does any evidence distinguish the favoured hypothesis from the ordinary rival, or does it merely fail to rule it out?
• Would the arguer accept an equally elaborate story from someone arguing the other side of this question?

Legitimate when: An initially implausible hypothesis is properly preferred when it earns the preference on evidence rather than on mere compatibility: when the ordinary rivals have been checked and positively fail, when the surprising hypothesis makes risky predictions that then come true, or when it explains further facts it was not invented to cover. Continental drift, Helicobacter pylori causing ulcers, and the meteorite impact at the end of the Cretaceous were all the least plausible hypothesis on the table until independent evidence made them the best one.

Ancient: Aristotle argues that where two demonstrations reach the same result, the better one is that which depends on fewer postulates or hypotheses, on the ground that the additional assumptions are so much extra that must itself be established. He is discussing the comparative worth of demonstrations rather than naming any fallacy, but this is the earliest clean statement of the parsimony standard that the far-fetched hypothesis violates. — Aristotle, Posterior Analytics I.25 (86a33-35)

Built from: Logically Fallacious — Least Plausible Hypothesis · Stanford Encyclopedia of Philosophy — Abduction (Inference to the Best Explanation) · Stanford Encyclopedia of Philosophy — Simplicity · Wikipedia — Occam's razor

Limited Depth

Explaining something by naming the group it belongs to, when belonging to that group just means doing the thing you were asked about.

In a pediatrician's waiting room, Nadia mentions that her son screams the whole way through the grocery store every single week. The woman next to her nods and says, "Well, he's two. That's what twos do." Nadia says thanks and feels briefly reassured, and then, in the car, realizes she knows nothing she didn't know before — not whether it's the fluorescent lights, the missed nap, the timing after lunch, or the shopping cart. Being two was never the thing she could change; it was just a nicer-sounding way of saying that a two-year-old was screaming.

The tell: Something got answered and yet you can't act on it. Say the answer out loud and see whether it still points at anything once you drop the category name — if what's left is "it does that because it's the sort of thing that does that," you've been handed a label instead of a cause. The dead giveaway is that the answer arrives before anyone has looked into anything.

The move: A phenomenon P is to be explained. The arguer offers as its explanation the fact that the subject belongs to a category K, where membership in K is constituted by, or simply named after, doing P — 'the dog gets into the bin because he is a dog', 'she is generous because she has a generous nature'. The explanans cites no mechanism, no antecedent condition, and no factor that could have been established independently of P itself; it relabels P at one level of abstraction up and stops.

Why it fails: An explanation earns its keep by citing something that could have been known independently of the thing explained, and that constrains what else would have to be true — a mechanism, a cause, a law, a disposition with its own detectable signature. Category membership fixed by the very behaviour in question supplies none of that: it cannot be checked except by observing P, it forbids nothing, and it would have been offered whatever P turned out to be. This is Molière's dormitive virtue, the doctor who explains that opium induces sleep because it possesses a sleep-inducing power. The account is not false — dogs do get into bins — but the causal information content is zero, and because it has the grammar of an explanation it closes inquiry that has not actually been conducted. That last effect is the real damage: shallow explanations do not merely fail to inform, they persuade people that the question has been answered.

• Could the cited factor have been established without already observing the thing it is supposed to explain?
• Does the explanation rule anything out — is there an observation it forbids, or would it have been given no matter how the case turned out?
• Is a mechanism named that connects the cited factor to the effect, or only a category, a trait-name, or a nature?
• Strip out the category label and restate the claim: does anything remain beyond 'P happens because this is the kind of thing that does P'?

Legitimate when: Citing a category is a legitimate shorthand when the category carries independently established content — saying a patient's fever is from influenza, or that a bridge failed because the steel was fatigued, points to a body of mechanism that could be spelled out on request and that predicts other things. It is also fine as a deliberate stopping point when everyone agrees no deeper account is wanted: 'I need oxygen because I'm human' is a shorthand, not an explanation, and nobody is misled. The fallacy begins when the label is doing no work beyond renaming the effect, and inquiry is closed on that basis.

Built from: Logically Fallacious — Limited Depth · Logically Fallacious — Limited Scope · Stanford Encyclopedia of Philosophy — Scientific Explanation · Wikipedia — Circular definition

Limited Scope

An explanation built to cover one case and nothing else — often just the same fact said back to you in different words.

In the kitchen at Ruta's, the soufflé comes out of the oven flat for the third night running. The sous chef tells the line, "It fell because it didn't hold its rise," and goes back to plating. Marisol, on garde manger, waits a beat and asks whether anything would come out different tomorrow if they believed that — the oven door, the egg whites, the copper bowl, the humidity off the dish pit. Nobody has an answer, because what they were given was the flat soufflé handed back to them wearing a hat.

The tell: You feel the conversation close without anything being learned, and if you try to use the answer for anything — to fix it, to predict tomorrow, to bet on it — there's nothing to grab. Ask what else it explains and watch the silence. Very often you'll find the words in the answer are just the words in the question, rearranged.

The move: A phenomenon P is to be explained. A theory T is offered which accounts for P and for nothing else: T was constructed to cover P, its terms are fixed by P, and it issues no consequence bearing on any case other than P. The arguer treats the fact that T fits P as support for T, though fitting P is the only thing T was ever built to do.

Why it fails: A theory gets confirmed by surviving the chances it takes — by covering cases it was not designed around, by forbidding outcomes that then fail to occur, by yielding predictions someone can go and check. A theory whose reach is exactly one datum takes no such chances, so the datum cannot raise its credibility above the level of the bare stipulation that produced it; it has been fitted, not tested. Scope is one of the standard criteria of adequacy for explanations precisely because breadth is what converts fit into evidence. In the narrowest version the explanation merely restates the phenomenon in different vocabulary — 'the car broke down because it stopped working', 'they decided hastily because they didn't take enough time' — and then not even one new fact has been added. The persuasive damage is that a one-case theory has the shape of an answer, so it discourages the search for an account that would also cover the next case.

• Name one thing besides the case at hand that this explanation covers or predicts. If nothing can be named, scope is zero.
• Does the explanation forbid any observation — is there something that, if seen, would count against it?
• Was the explanation constructed after the fact specifically to fit this case, using terms defined by this case?
• Restate the explanation without the words from the phenomenon itself: does any new content survive, or is it the same claim in other words?

Legitimate when: Genuinely singular events do have singular explanations, and there is nothing wrong with them: this bridge fell because this weld failed, and that account need not cover other bridges to be correct — but it earns its standing by pointing at independently checkable facts about the weld, not by fitting the collapse. Restatement is also legitimate when it is announced as such: replacing 'because' with 'in other words' turns a failed explanation into an honest clarification, and no one is misled.

Built from: Logically Fallacious — Limited Scope · Logically Fallacious — Limited Depth · Wikipedia — Ad hoc hypothesis · Stanford Encyclopedia of Philosophy — Scientific Explanation

Logic Chopping

Picking at a tiny wording detail that wouldn't change the answer either way, so the real question never gets dealt with.

Scout troop planning meeting, eight o'clock on a Wednesday, and Alice proposes that anyone driving kids to the campout should have a working spare tire. Rob raises his hand: what counts as "working", exactly — a donut? What about run-flats, which have no spare at all? And is a person who drives one kid a "driver" in the relevant sense, or does the rule mean the caravan lead? Twenty minutes later the group has not voted on anything, and Rob has still not said whether he thinks the checks are a good idea.

The tell: You keep granting the point and it keeps not helping. Each objection is small and often technically right, and yet the thing you actually came to settle is drifting further away. Notice whether the other person has staked out any position of their own that could be pinned down the same way — usually they haven't, and that asymmetry is the whole game.

The move: A claim or request C is put forward on some substantive question Q. Rather than engaging C, the arguer raises an objection O aimed at a peripheral feature of how C was worded, classified, or qualified — a borderline case, an imprecise term, an unstated caveat. O may well be correct. But O is immaterial: whether O is granted or denied, the standing of C on Q is unchanged. The arguer nonetheless treats the unresolved O as blocking any resolution of Q.

Why it fails: Relevance in argument is a matter of whether settling a point would move the question, and the quibble is chosen precisely because settling it would not. Since the fine distinction leaves the substantive claim exactly where it was, granting it costs the arguer nothing and gains the challenger nothing — which is why the manoeuvre can be repeated indefinitely and functions as a delay rather than a reply. Being correct is no defence here: the defect is not falsity but immateriality, which is what separates logic-chopping from an ordinary good objection. Two further features usually give it away. The demand for precision is applied asymmetrically, against one side's wording only, while the chopper's own terms go unexamined. And the objection attacks the letter of what was said rather than the strongest reading of what was meant, so the claim actually on the table is never confronted. The technical apparatus of logic is being used as a shield rather than an instrument.

• Grant the objection outright, for the sake of argument: does the substantive question now come out differently? If not, the objection is immaterial.
• Is the same standard of precision being demanded of the objector's own claims, or only of the other side's?
• Does the objection engage the strongest reasonable reading of the claim, or only its exact wording?
• Has the objection been raised in place of any position of the objector's own — is there anything they are prepared to be pinned to?

Legitimate when: Precision is often exactly what a dispute needs, and demanding it is legitimate whenever the imprecision is load-bearing: if the argument turns on the meaning of 'reasonable', or 'income', or 'first offence', then pinning the term down is the substantive work, not an evasion of it. Contract drafting, statutory interpretation, and definitions in science all live here. The test is not how fine the distinction is but whether settling it would change the answer to the question actually being argued.

Ancient: The eristic brothers Euthydemus and Dionysodorus trap their interlocutors with captious verbal quibbles rather than argument — most famously reasoning that since Ctesippus has a dog, and the dog is a father, the dog is Ctesippus's father, so Ctesippus beats his own father. Socrates' complaint is not that the wordplay is easy to refute but that it is designed to trip an opponent rather than to settle the question, which is exactly the charge behind the modern label. — Plato, Euthydemus

Built from: Logically Fallacious — Logic Chopping · Wikipedia — Trivial objections · Wikipedia — Euthydemus (dialogue) · Stanford Encyclopedia of Philosophy — Fallacies

Logical Fallacies Listing with Definitions and Detailed Examples

Filing the label on the outside of the box as though it were one of the things inside the box.

Ms. Okonjo's ninth-graders type up the class survey results, and one of them enters the header row into the spreadsheet along with everything else. By Friday the summary slide reports that the most common favourite lunch is "Favourite Lunch," chosen by one student. Everyone laughs, and then a boy in the back asks whether they should tell the office, because the office already used the numbers for the cafeteria order. The header was never a lunch; it was the word for where the lunches were written down.

The tell: You go to look something up, find the entry, and discover there's nothing in it you could ever use — no way to tell whether a given case counts, no way to be wrong about it. It reads like a title rather than a thing. If you try to explain it to someone, you'll find yourself describing a webpage instead of describing anything a person does.

The move: A collection is assembled by harvesting items from documents. A string that belongs to a document's furniture — its title, a navigation heading, a section label — is captured by the same rule that captures genuine entries, and is then filed as though it were an entry of the same kind. Downstream, arguers reason from the collection as if every row denoted a real member of the category: 'it is in the list of X, therefore it is an X.' The label names no argumentative move; it names the page the moves were copied from.

Why it fails: The defect is a category error made by the collection rather than by any arguer inside it, so there is no argumentative move to diagnose here at all. What can be said is why the resulting row is dangerous: a catalogue's authority comes from its rows having been individually established, and a row produced by the harvesting rule rather than by scholarship carries no such warrant, yet is indistinguishable in form from rows that do. Anyone reasoning from membership therefore inherits an unearned entitlement, and the error propagates every time the list is copied. 'Logical Fallacies: Listing with Definitions and Detailed Examples' is on its face the title of a reference page, not the name of a reasoning error: there is no pattern of argument it picks out, no condition under which an argument would count as committing it, and no literature that treats it as a move. It should be retired from the corpus, or kept only as provenance metadata pointing at the source page the real entries came from.

• Does the label describe something an arguer could do in an argument, or does it describe a document — a title, a heading, a table of contents?
• Can you state a condition under which a passage would count as committing it? If no such condition can be written, it is not an entry.
• Does any independent reference work treat this string as the name of a reasoning error, or does it appear only as the heading of a page that lists them?
• Is anything substantive recoverable from the source page that belongs in a genuine record before this row is retired?

Legitimate when: Reasoning from a catalogue is perfectly sound when the catalogue's rows have been individually vetted and the row you are relying on is one of them — that is what reference works are for. It is also legitimate to keep provenance rows in a dataset, recording which source a batch of entries came from, provided they are stored as metadata and never presented alongside the entries as though they were entries themselves.

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies · Wikipedia — List of fallacies · Wikipedia — Category mistake

Ludic Fallacy

Treating the messy real world as if it played by the tidy, spelled-out rules of a board game, so anything the game can't represent quietly stops existing.

Six months out, Priyanka builds her wedding day as a spreadsheet: every block timed to the minute, buffers costed, three weather scenarios weighted by thirty years of records for that county. She tells her sister the day is solved. On the day, the officiant's car is rear-ended on the highway, the venue's power goes at four, and her father-in-law announces an unscheduled forty-minute speech. None of the three was in the sheet — not because she got the odds wrong, but because the sheet had rows only for the things she had already thought of.

The tell: The numbers are suspiciously exact for a situation that is obviously not exact. When you ask what happens if something not on the list occurs, you get an answer about how unlikely the things on the list are — which is a different question. There's a warm, false calm to it: the more decimal places, the safer you feel, and nothing about the world has actually changed.

The move: A real-world domain D is modelled by a game-like structure G whose sample space, rules, and probability distribution are fully specified in advance — dice, cards, a roulette wheel, a Monte Carlo simulation with fixed parameters. Results computed inside G are then asserted about D as though the two were the same object. The step that is never taken is the one that would license the transfer: no argument is given that D's set of possible outcomes is bounded by G's, or that D's frequencies obey G's assumed distribution.

Why it fails: Games are constructed so that their uncertainty is complete and known: the space of outcomes is closed, the odds are stipulated, and nothing outside the rules can occur. Real domains have none of these properties — their outcome space is open, the distribution is estimated from a past that may not resemble the future, and the parameters themselves are uncertain. Computing precisely inside such a model quantifies only variance within the model and leaves out error of the model, which in open domains is usually the larger term. The consequences that would matter most are exactly the ones the model has no way to represent, so a run of ten thousand simulations returning no failure is evidence about the simulation, not about the world. Precision then does active harm: the tidy number displaces the vaguer but truer statement that the risk is not known, and confidence rises as the modelled range narrows, which is the opposite of what the epistemic situation warrants. Nassim Nicholas Taleb, who named the move in The Black Swan (2007), calls it 'basing studies of chance on the narrow world of games and dice', a special case of what he calls Platonicity — preferring the clean map to the messy country.

• Does the model's set of possible outcomes include the events that would matter most if they happened, or are those simply outside it?
• Where did the assumed distribution come from — is there evidence that real frequencies in this domain obey it, or was it chosen for tractability?
• Is the uncertainty being quantified variance inside the model, or uncertainty about whether the model is right? Is the second reported at all?
• If the model's own parameters were off by a factor of two, how much would the conclusion move?

Legitimate when: Formal probability is exactly the right tool where the domain really is closed and stationary: casino games, well-characterised physical processes, sampling from a known population, insurance lines with long stable loss histories. It is also legitimate in open domains provided the output is used as a lower bound or a scenario rather than as the answer — stress tests, sensitivity analysis, and explicitly reported model risk keep the same machinery honest by admitting what lies outside it.

Built from: Wikipedia — Ludic fallacy · Wikipedia — Black swan theory · Stanford Encyclopedia of Philosophy — Risk · Stanford Encyclopedia of Philosophy — Interpretations of Probability

Lying with Statistics

Every number is true and the picture they add up to is false, because somebody chose which numbers to show.

The gym on Fulton Street papers its windows with a poster: members lose an average of eighteen pounds in twelve weeks. Ruth signs up, and four months later mentions to a trainer that she's nowhere near it. He tells her, off the record, that the eighteen pounds is the average among people who finished all twelve weeks and logged every session — about one in nine who join. The number is real. It just isn't a number about people like Ruth, and the poster never said which people it was about.

The tell: The figure is very specific and the story around it is very vague — you know the number to a decimal place but not who was counted, over what stretch, or out of how many. When you ask, the answer changes subject to how impressive the number is. If you catch yourself thinking "that seems like a strangely chosen window of time," trust it; that's usually exactly what happened.

The move: A numerical claim N is presented in support of a conclusion C. Every individual figure in N is accurate, and the presentation is technically defensible. But at least one of the choices that produced N — which interval was measured, which subgroup was counted, which denominator was used, where the axis begins, which of several averages is called 'the average', how the categories were binned — was made because it favours C, and that choice is not disclosed. The audience evaluates N as though it were the neutral summary of the underlying data, which it is not.

Why it fails: A statistic supports a conclusion only under a description of how it was produced; strip that away and the number is uninterpretable, because for most datasets there exists some defensible-looking summary favouring almost any conclusion you like. The deception works precisely because nothing false is stated — the manipulation lives in the selection, and selection is invisible in the output. Since the same raw data would have yielded a very different figure under a neutral protocol, the audience's confidence is being drawn from the choice rather than from the world. This is also why the move is so hard to rebut in real time: refuting it requires reconstructing the whole series, while committing it requires one sentence. The label is an umbrella, taken from the title of Darrell Huff's How to Lie with Statistics (1954), and it covers a family of distinct defects — truncated or 'gee-whiz' axes, cherry-picked endpoints, biased sampling, missing denominators, the semi-attached figure, mean-versus-median switching, improper aggregation. Any live diagnosis should name which of these is present, since the umbrella term names no single failure.

• Which specific defect is present: truncated axis, cherry-picked start and end dates, biased or self-selected sample, missing denominator, switched average, or regrouped categories? Name it.
• What does the full series look like with the axis from zero and every period included — does the effect survive?
• Who chose the comparison point, and would a neutral party working from the same raw data have chosen it?
• Is a percentage given without the base it is a percentage of, or a count given without the population it came from?

Legitimate when: Summarising is not distorting: every honest statistic involves choices about interval, grouping, and measure, and those choices are legitimate when they are disclosed, defensible on grounds independent of the conclusion, and stable — the same protocol applied to unfavourable data would still have been reported. Truncating an axis is fine when small movements are the subject and the truncation is labelled; a subgroup figure is fine when the subgroup was specified before the data were seen.

Built from: Wikipedia — Misuse of statistics · Wikipedia — How to Lie with Statistics · Wikipedia — Base rate fallacy · Stanford Encyclopedia of Philosophy — Fallacies

Magical Thinking

Believing that a ritual, a word, or an object reaches out and changes something it has no way of touching.

Thursday night bowling league at Sunrise Lanes, and Denny will not roll a frame without the faded orange shirt he wore the night he shot 268. When it's in the wash he begs off. His teammate Roz finally pulls the score sheets from the last two seasons and lays them out: he averages 176 in the shirt and 174 out of it, and he has bowled his three worst games of the year in it. Denny looks at the sheets for a while and says the shirt must have been in the dryer too long.

The tell: You notice you're keeping score in only one direction — the times it worked are vivid and the times it didn't have quietly gone missing. Ask yourself what would have to happen for you to give the ritual up, and if the honest answer is nothing, that's the thing to look at. There's usually a little spike of dread at the thought of skipping it, and the dread arrives long before any reason does.

The move: An act, thought, word, or object R stands in some non-causal relation to an outcome O — R resembles O, R has touched something connected to O, R was performed shortly before O, or R symbolically represents O. From this the arguer concludes that R exerts causal influence over O, and reasons practically on that basis: perform R to bring O about, or avoid R to prevent it. No mechanism connecting R to O is proposed, and the relation actually cited — resemblance, contact, symbolism, or mere temporal adjacency — is one that carries no causal implication.

Why it fails: Resemblance and contiguity are relations among our representations, not among events in the world, and the argument treats a feature of how we group things as if it were a feature of how they act on one another. J. G. Frazer set out the two standing patterns as the law of similarity, that like produces like, and the law of contagion, that things once in contact continue to act on one another; both take a principle of association and read it out onto nature. Because no mechanism is proposed, there is nothing that could be checked, adjusted, or bounded, so the claim makes no contact with evidence in either direction. The supporting record is also gathered improperly: the hits are counted and remembered while the misses are not, and where R is performed constantly there is no comparison case in which R was omitted, so the covariation that would be needed to establish anything has never been observed. And the practice tends to be self-sealing — success confirms the ritual, failure shows it was performed wrongly or not enough — which converts a causal claim into one that no outcome can disturb.

• What physical, biological, or institutional mechanism could carry influence from the act to the outcome? Can any be stated?
• Is the cited connection actually resemblance, prior contact, symbolism, or mere sequence in time — none of which is a causal relation?
• Does the claimed pattern survive counting the times the act was performed and the outcome did not follow?
• Is there any case on record in which the act was omitted, so that the two situations could be compared at all?
• When the outcome fails to arrive, is the ritual revised or is the failure blamed on the performance? If the latter, no evidence can reach the claim.

Legitimate when: Rituals can genuinely produce effects by ordinary routes, and saying so is not magical thinking: a pre-serve routine steadies a tennis player's attention, a team chant raises effort, a lucky shirt reduces anxiety and so improves performance, and a wedding vow really does change a legal status because an institution makes it so. All of these name a mechanism and predict where the effect will and will not appear. Reasoning from a well-documented correlation to a probable cause is likewise fine when the correlation has been measured with the misses included and rivals ruled out.

Ancient: Theophrastus sketches a man for whom chance events are potent causes demanding countermeasures: a weasel crossing his path stops him until someone else has gone by or he has thrown three stones over the road, and he purifies himself over dreams, owls, and stumbling. The sketch is a character portrait rather than an argument, but it is a precise ancient record of treating resemblance and coincidence as causal power, with no mechanism anywhere in view. — Theophrastus, Characters XVI ('The Superstitious Man')

Built from: Wikipedia — Magical thinking · Wikipedia — Sympathetic magic · Logically Fallacious — Magical Thinking · Wikipedia — Superstition

McNamara Fallacy

Deciding that whatever you managed to count is the whole story, and that anything you couldn't count doesn't count.

At the Tuesday floor meeting of a call centre, the operations manager puts up the wall board: average handle time down to four minutes twelve, calls per hour up, every target green. A senior rep raises her hand and says people are hanging up angry and calling back three days later about the same thing. He says that's a feeling, not a metric, and moves on. Nobody in the room disputes his numbers — the repeat calls are simply counted as fresh calls, so the board goes greener the worse it gets.

The tell: You raise something everyone in the room knows is true and get told it isn't data. Watch for the slide from "we can't measure that" to "that doesn't matter" to "that isn't real" — it usually happens across three sentences and nobody marks the transitions. The feeling is of being asked to disbelieve your own eyes on the authority of a spreadsheet that never looked at what your eyes were looking at.

The move: A question Q about some domain D is to be settled. The arguer restricts admissible evidence to the quantified subset M of facts about D, on the ground that only M is data. Facts in D that resist measurement are first set aside as unusable, then treated as unimportant because they are absent from the analysis, and finally denied altogether. A conclusion about D is then drawn from M as though M were D, with no argument that the measured subset is representative of the whole.

Why it fails: The move is not a preference for rigour but a substitution: an argument about the world is replaced by an argument about the instrument, and the gap between them is never bridged. Measurability is not distributed at random across the facts that bear on a question — what gets counted is what was cheap or convenient to count, which makes M a systematically biased sample of D rather than a merely smaller one. So the usual defence, that partial evidence is better than none, does not apply; biased evidence treated as representative can point the wrong way with more confidence than no evidence would. There is also a feedback effect: once M is the criterion of success, effort migrates toward moving M rather than toward the underlying goal, so M degrades as an indicator of D precisely because it is being used as one. Daniel Yankelovich set out the progression in 1971 as four steps — measure what is easily measured; disregard or assign an arbitrary value to what is not; presume that what cannot be measured easily is not important; and finally say that what cannot be easily measured does not exist — naming it for Robert McNamara, whose Vietnam-era body counts stood in for the state of a war they did not describe.

• Name a factor that clearly bears on this question but is not in the numbers. Where does it enter the analysis — or does it enter at all?
• Were these particular metrics chosen because they capture what matters, or because they were the ones available to collect?
• Is unmeasured evidence being weighed and outweighed, or is it being ruled inadmissible before the weighing starts?
• Since the metric became the target, has it moved in ways that plainly did not track the underlying thing it was supposed to indicate?
• Does the quantified subset differ systematically from the whole it is standing in for, or is there reason to think it representative?

Legitimate when: Insisting on measurement is usually right, and the corrective for bad decisions is very often better data rather than more anecdote. The shape is sound when the metrics have been validated against the outcome they proxy for, when their known blind spots are stated and carried alongside the numbers, and when unmeasured considerations are argued about rather than ruled out — a hospital that tracks readmission rates while explicitly discussing what readmissions miss is doing this correctly. The fallacy begins at the third and fourth of Yankelovich's steps, not the first.

Built from: Wikipedia — McNamara fallacy · Wikipedia — Goodhart's law · Wikipedia — Misuse of statistics · Stanford Encyclopedia of Philosophy — Measurement in Science

Meaningless Question

Demanding an answer to something that isn't really asking anything, then treating whatever comes out as a real answer.

The wellness app Jonah's employer rolled out ends its Sunday check-in with a slider: "How aligned is your energy with the year ahead?" — nought to a hundred. He picks 62 because it seems modest. In January his manager pulls up the dashboard and notes with concern that Jonah's alignment has fallen eleven points since autumn, and asks what's been going on. Jonah has no idea; there was never anything the number was about, and now it's a topic in his review.

The tell: You find yourself hunting for the answer and realize you're actually hunting for what the question could possibly mean. The words are all ordinary and the sentence is shaped right, so you assume the fault is yours. Watch for the moment where you'd have to make something up in order to reply — and for the fact that once you reply, whatever you said starts being treated as a thing you believe.

The move: A question Q is posed and an answer demanded. Q has the surface grammar of a well-formed question, but it applies a predicate outside the range where the predicate has application — asking what lies north of the North Pole, what happened before time, or how heavy a colour is. Because no state of the world would make one answer correct rather than another, any answer given, and any scoring of the answers, presupposes that Q has a determinate sense. That presupposition is the smuggled claim, and it is false.

Why it fails: A question is answerable only if its terms have application in the domain it is asked about; where they do not, the question does not have a false answer, it has no answer, and the grammar is misleading everyone. This is Gilbert Ryle's category mistake wearing an interrogative: 'north of' is defined only within a coordinate system that terminates at the pole, so the string is well formed and empty. The argumentative damage comes from what follows. Demanding an answer forces the respondent to accept the presupposition merely by speaking, exactly as a complex question does, and once an answer is on the record it can be treated as a commitment and argued against — so the questioner gains a position the respondent never held. Answers can then be compared, ranked, and scored, producing the full appearance of a settled dispute about nothing. The essential discipline is to separate this from questions that are merely hard, currently unanswerable, or contested in their framing: those have determinate senses and we simply lack the means or the agreement, whereas here there is nothing to lack.

• Name one observation or finding that would favour one answer over another. If nothing can even in principle be named, the question has no sense rather than an unknown answer.
• Are the key terms being applied to a domain where they have no application — spatial words to non-spatial things, temporal words to what precedes time, quantities to things that do not admit of amount?
• Can the terms be given definite content without someone simply stipulating it on the spot?
• Is this genuinely senseless, or merely very difficult, undecided, or dependent on a framework the parties have not agreed? Only the first is the fallacy.
• Does answering at all commit the respondent to something they have not accepted?

Legitimate when: Hard questions, open questions, and questions whose answers we may never obtain are all perfectly meaningful, and pressing them is how inquiry works — what caused the Bronze Age collapse, whether there is life elsewhere, what the first speakers of a lost language believed. Questions inside a stipulated framework are also fine once the framework is stated: 'how many angels fit on a pinhead' becomes tractable the moment someone fixes what angels are and whether they occupy space. Framework-relative and metaphorical questions are legitimate provided nobody pretends the answers are findings about the world.

Built from: Logically Fallacious — Meaningless Question · Wikipedia — Category mistake · Wikipedia — Not even wrong · Stanford Encyclopedia of Philosophy — Questions

Methodology

Counting the page that explains how a list was made as though it were one of the things on the list.

Marguerite scans her grandmother's recipe binder page by page so the cousins can all have it. When the family cookbook comes back from the printer, between the pierogi and the poppyseed roll there is a full recipe titled "How To Use This Binder", with no ingredients and no oven temperature — just her grandmother's note about keeping the sweet ones at the back. It's the most touching page in the book and it is not a recipe, and if anyone tried to cook it they would find there was nothing there to cook.

The tell: You look up an entry and there's nothing in it you could ever apply — no way to tell whether a given case counts, nothing you could be wrong about. It reads like a heading someone forgot to take out. If explaining it to a friend means describing how a document was organized rather than describing something people do, that's the sign.

The move: A corpus is built by extracting entries from source documents. A structural heading — here the word 'Methodology', which introduced a source document's account of how its list was compiled — is captured by the extraction rule alongside genuine entries and stored as one of them. Readers then treat the row as denoting a reasoning error, and may reason from its presence: 'the catalogue contains it, so there is such a move.' The label picks out a section of prose about how the catalogue was made, not anything an arguer can do.

Why it fails: There is no argumentative move here to critique, so the honest account is of the cataloguing failure that produced the row and of the harm it does. A reference work's rows carry authority because each was established separately; a row generated by the scraper rather than by an editor carries none, yet sits in the same table, in the same typeface, indistinguishable from rows that were earned. That is why such artifacts are worse than merely useless — they lend the catalogue's credit to something that was never assessed, and every downstream copy inherits it. In this instance the underlying prose may well have been worth keeping: a source's methodology section describes how its entries were selected and what its inclusion criteria were, which is exactly the provenance information a research corpus should hold. The remedy is therefore not deletion alone but relocation — move any real methodological content to collection-level documentation, and retire the row from the entries. There is one adjacent defect the label could be stretched to name, the practice of attacking a study's methods in the abstract without identifying any error that would change its result, but that is properly logic chopping or a red herring and already has its records.

• Does the label name something an arguer could do in an argument, or does it name a section of a document about how a list was assembled?
• Can you state a test that would decide whether a given passage commits it? If no such test can be written, this is not an entry.
• Does any independent reference work treat this word as the name of a reasoning error, or does it appear only as a heading?
• Is there methodological prose behind this row worth preserving in collection-level documentation before the row is retired?

Legitimate when: Methodological criticism is entirely legitimate and is often the strongest objection available: challenging a sample, a control, a coding scheme, or a measurement instrument is real argument whenever the challenger can say how the flaw would change the result. Keeping methodology notes in a corpus is likewise good practice — the error is only in storing them as though they were entries of the kind the corpus catalogues.

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies · Stanford Encyclopedia of Philosophy — Scientific Method · Wikipedia — Methodology

Misleading Vividness

Letting one unforgettable story outweigh a mountain of boring numbers that say it almost never happens.

Gate B14, an hour before the flight to Tessa's wedding, and Aunt Corinne is refusing to board. She watched a documentary in March about an engine failure over the Atlantic — the oxygen masks, the man who called his daughter. Tessa's brother pulls up the numbers on his phone and reads out how many flights land safely for every one that doesn't. Corinne says she knows all that, and she can still hear the recording of the pilot. She ends up driving eleven hours through the mountains, on the leg of the trip that is actually dangerous.

The tell: One story keeps replaying in your head and the numbers slide off it. You'll notice you can picture the story in detail — the faces, the sounds — and cannot picture the thousands of ordinary cases at all, because ordinary cases have nothing to picture. If someone told you an equally gripping story pointing the other way, and you know it wouldn't change your mind, then the gripping part was never the reason.

The move: A population is characterised by a body of systematic evidence showing that most members are not F. One or a few cases c are then presented which are F and which are described in dramatic, concrete, emotionally charged detail. The arguer concludes that F should be expected of the population, or acts as though the rate of F were far higher than the record shows. The vividness of c, rather than any argument that c is representative or that the systematic evidence is flawed, is what carries the audience from the case to the claim.

Why it fails: A sample supports a generalisation in proportion to how it was drawn, and the cases that reach us vividly were not drawn at random — they reached us because they were unusual, which is the property that makes them newsworthy, memorable, and repeatable. So salience is inversely related to representativeness in exactly the situations where this move gets used, and the feature doing the persuading is evidence of the opposite of what it is taken to show. The systematic evidence is not rebutted, either: it is simply outweighed by something with no evidential standing, since one dramatic case is fully consistent with a very low rate and does nothing to raise the estimate. The psychology behind it is well documented as the availability heuristic — ease of recall stands in for frequency — and the argumentative test is symmetry: an equally vivid case running the other way could be produced for almost any population, and if that one would not move the conclusion, then vividness was never functioning as evidence at all. There is a legitimate residue worth preserving: a single case can rebut a universal claim, and can raise a hypothesis worth investigating, but neither of those is a claim about how common the thing is.

• Is the vivid case representative, or is it memorable precisely because it is rare? How did it come to your attention?
• What does the systematic evidence say about the rate, and what argument has been given for setting it aside rather than merely feeling past it?
• Could an equally vivid case be produced for the opposite conclusion? If so, and if that one would not persuade, vividness is not doing evidential work here.
• Is the case being used to refute a claim that something never happens — which one case can do — or to establish how often it happens, which one case cannot?

Legitimate when: A single case is decisive against a universal claim: one working counterexample refutes 'this never fails', and one documented harm refutes 'this is entirely safe'. Vivid cases are also legitimate for raising a hypothesis, for illustrating a pattern the statistics have already established, and for conveying what a rate means in human terms. And when the base-rate evidence is itself known to be bad — undercounted, self-reported, collected by an interested party — a well-documented case may properly outweigh it, but then the argument runs through the defect in the data, not through the drama of the story.

Built from: Logically Fallacious — Misleading Vividness · Wikipedia — Anecdotal evidence · Wikipedia — Availability heuristic · The Fallacy Files — Hasty Generalization · Wikipedia — Base rate fallacy

Missing Data Fallacy

Insisting that the evidence which would prove you right is out there somewhere, and treating that as if it were the same as actually having it.

Deon has been telling everyone at the climbing gym that a collagen powder is what healed his shoulder. Priya, who works at a physical therapy clinic, pulls up a summary of six trials that found no effect at all. Deon barely looks at it. "Those used the wrong kind," he says. "The study that would show it properly just hasn't been run yet." A month later she brings him a trial that used exactly the kind he takes, and he says the dose was too low — the right study is still out there, always one study away.

The tell: You hand over the very thing they asked for and nothing lands. The goalposts don't move, exactly; the game just gets postponed. After a while you notice that whatever you bring, the winning card is always still somewhere in the deck, and you are the only person in the conversation being asked to produce anything.

The move: An arguer A holds a claim H. Evidence E is produced that H cannot accommodate. Rather than revise or withdraw H, A posits a body of data D that has not been observed, stipulates that D is exactly what H would need in order to survive E, and treats H as still standing on the strength of D. D is characterized entirely by the work it would do for H: no independent reason is given for thinking D exists, no custodian or location is named, and no route to producing D is offered.

Why it fails: Support has to come from data that exists, not from data whose only described property is that it would rescue the claim if it existed. The move takes a debt — the evidence that is absent — and books it as an asset. Because D is defined by the need it fills rather than by any observation, the identical maneuver is available to every claim whatever, including the direct denial of H; a device that defends everything discriminates nothing and therefore confers no support on anything. It also removes H from the reach of evidence in practice: any adverse result can be absorbed as further proof that the real data has not come in yet, so no possible finding counts against it. What is presented as patience about incomplete science is in fact a standing exemption from being tested.

• Has the missing data been named concretely — who holds it, where it sits, how it could be obtained — or is it described only by what it would show?
• Was the appeal to unavailable data made before the objection landed, or only after it landed?
• Would exactly the same appeal work just as well for the opposite claim, or for any rival claim in the room?
• Can the arguer name a result they would accept as settling the matter against them, or can every finding be absorbed as 'the real data isn't in yet'?

Legitimate when: Pointing at evidence you do not currently have in hand is perfectly legitimate when that evidence is independently known to exist and can be produced: a registered trial whose results are due on a stated date, a paper you remember but must look up, an audit already underway. The line is whether the absent data has an identity apart from the rescue it would perform — a custodian, a location, a timetable — and whether the arguer treats the claim as unsettled while waiting rather than as vindicated in advance.

Built from: Logically Fallacious — Missing Data Fallacy · Wikipedia — Argument from ignorance · Wikipedia — Survivorship bias

Modal (Scope) Fallacy

also known as: modal scope fallacy

Turning "this had to follow from that" into "this had to happen no matter what" — sliding the word 'must' one step too far down the sentence.

Tomás's grandmother told him in September that he would drop out of the paramedic program by spring, and in March he did. At the kitchen table he shrugs. "She knew. So it was always going to happen. There was nothing I could have done." His sister points out that Abuela also called rain for Thursday and it rained on Thursday, and nobody thinks Abuela made it rain. All her record shows is that when she says a thing, it turns out that way — not that Tomás's hands were tied to the wheel.

The tell: The reasoning sounds airtight and somehow leaves you with nothing left to do. You can't find the crack, but you notice you have just been talked out of your own steering wheel, and that the same words would work equally well for absolutely anything that happened to turn out the way it did.

The move: A conditional is established as necessary — necessarily, if P then Q — where the necessity governs the whole link between P and Q. The arguer then adds that P holds and draws not merely Q but the necessity of Q: from box(P then Q) together with P, they take box Q. The necessity operator has been detached from the conditional as a whole and reattached to its second half alone. The same slide appears in de dicto / de re dress: 'necessarily, every bachelor is unmarried' is read as 'of every bachelor it is necessary that he be unmarried', and from there that no bachelor can marry.

Why it fails: Necessity of the consequence and necessity of the consequent are two different properties, and only the first has been established. Box(P then Q) says that no situation contains P without Q; it says nothing whatever about whether Q could have failed to obtain, because in the situations where Q fails, P fails too. To get box Q you need box P as well — the necessity has to be fed in from the premise side, not conjured out of the strength of the link. When P is merely contingent, Q inherits that contingency: the connection holds unbreakably and both ends of it were free to have been otherwise. Read the other way, the argument would convert every guaranteed connection into a guaranteed outcome, which makes everything that follows from anything unavoidable — the mark of a step that proves far too much.

• Where does the word 'necessarily' or 'must' actually sit — in front of the whole if-then, or in front of its second half alone?
• Is the statement that supplies the starting fact itself necessary, or is it something that merely happens to be true?
• Can you describe a coherent situation in which the starting fact had not held? If you can, the outcome is not unavoidable.
• Does the argument lose all its force if you restate it as 'given that P, Q' instead of 'P, therefore Q had to happen'?

Legitimate when: The same shape is entirely legitimate when the starting condition is itself necessary: from box(P then Q) together with box P, box Q follows by ordinary modal reasoning. It is also fine when Q is necessary on independent grounds — a definition, an arithmetical truth — and it is fine as everyday shorthand, where 'then it must be Friday' means only 'given what we have agreed, it is Friday' and no claim about unavoidability is being smuggled in.

Ancient: Aristotle grants that what is, necessarily is when it is, while denying that whatever is, is therefore necessary without qualification — exactly the distinction this move collapses. Later readers sharpen the point against the fatalist argument the chapter examines: the Stanford Encyclopedia notes it 'looks, indeed, suspiciously like a fallacious move from "Of necessity, if it is true that a sea-battle takes place on 1/1/2100, a sea-battle takes place on 1/1/2100" to "If it is true that a sea-battle takes place on 1/1/2100, a sea-battle takes place of necessity".' — Aristotle, De Interpretatione 9 (the sea-battle chapter)

Built from: Wikipedia — Modal fallacy · Stanford Encyclopedia of Philosophy — Fatalism · Stanford Encyclopedia of Philosophy — Future Contingents

Moralistic Fallacy

also known as: ought-is fallacy

Deciding something can't be true because it would be terrible if it were.

At the end of practice Coach Whitaker mentions a study on kids who play one sport year-round and how often their shoulders and elbows give out. Gwen, one of the parents, shakes her head before he has finished. "If that were right, half these girls would have to quit their travel team, and it would break their hearts. So it can't be right." She hasn't asked who was in the study or how long they were followed. What she has done is measure the finding against how much she would hate its consequences, and found it wanting.

The tell: Notice that nothing in the pushback is about the evidence. You brought a claim about how things are and got back a claim about how bad that would be, delivered with all the force of a correction. There is a faint pressure to agree, because disagreeing starts to feel like siding with the bad outcome rather than like reading the study.

The move: P is a factual claim about how the world is. The arguer holds an evaluative claim about P: that P's being true would be morally bad, unjust, or dangerous, or that the world ought not to be a P-world. From that evaluative claim the arguer draws the factual one — that P is false. Run in the other direction, a morally desirable state of affairs Q is declared actual because it ought to be actual. Schematically: it ought to be that not-P; therefore not-P. Bernard Davis, who gave the label its current use in Nature in 1978, described it as in effect an illogical effort to derive an 'is' from an 'ought'.

Why it fails: The desirability of a state of affairs is not among the things that make it obtain. Moral evaluation takes the facts as its object; it does not supply them. An evaluative premise about how the world should be therefore leaves the question of how the world is exactly where it was, and the argument is missing the step that would license the crossing. There is a further cost beyond the missing step. Because the move rules out findings in advance of looking, it converts a moral commitment into a standing ban on inquiry, and the commitment then has to be defended by suppressing evidence rather than by being right about the world. Worse, it makes good moral positions hostage to contingent facts: if the finding turns out to hold after all, the moral claim that was made to rest on its falsity goes down with it.

• Is the reason offered against the claim a reason about the evidence, or a reason about what would follow if it were true?
• Strip out every word about harm, unfairness, or what people would do with the finding — is there anything left that bears on whether it is so?
• Would the speaker accept the same style of reasoning from somebody with the opposite moral commitment?
• Does the speaker's moral position actually require the factual claim to be false, or would it survive intact if the claim turned out to be true?

Legitimate when: An 'ought' can bear legitimately on an 'is' in two ways that are not this move. First, when the disputed claim is itself about values or obligations, moral reasons are simply on topic. Second, moral stakes can properly raise the evidential bar: a finding with grave consequences may reasonably be held to stricter standards of replication before anyone acts on it, provided the standard is applied symmetrically and can in principle be met. What is never legitimate is treating the badness of a result as a reason to score it false.

Built from: Wikipedia — Moralistic fallacy · Wikipedia — Naturalistic fallacy · Stanford Encyclopedia of Philosophy — Hume's Moral Philosophy

Moving the Goalposts

also known as: raising the bar

Someone keeps raising the bar every time you clear it, so you can never actually win.

Your manager says, "Get five 5-star reviews from customers and the raise is yours." You work hard and get five. She looks them over and says, "Well, those were probably easy customers — get me five from people who complained first." You go get those too. Now it's, "Online reviews don't really count, I need in-person feedback forms." The goal was never really five reviews. It was whatever you hadn't done yet.

The tell: That queasy, deflating feeling of "wait, that wasn't the deal we agreed on" — you did the actual thing that was asked, and instead of a yes, you got a brand-new, harder ask that nobody mentioned before. If you rewind the tape, the finish line only ever moves after you reach it, never before.

The move: A standard S is set (explicitly or by clear implication) as what would suffice to establish claim C. The evidence required by S is then produced and meets it. Instead of accepting C, S is quietly replaced with a stricter, previously unstated standard S′, and C is declared unproven under S′ — with no reason for the change other than the fact that S was just satisfied. The pattern repeats: whatever is produced to meet S′ triggers S″, and so on.

Why it fails: The move treats the burden of proof as infinitely adjustable rather than fixed in advance, so the claim is rendered unfalsifiable in practice: the accept/reject decision was never actually tracking the strength of the evidence, it was tracking whether the responder wanted to keep dodging. This differs from an honest change of mind because it is the criterion, not the evidence, that is being manipulated — and it is manipulated for the same reason every time: because the prior criterion just succeeded. Structurally it is a covert, unilateral redefinition of what counts as 'enough,' introduced after the fact and applied only in the direction that keeps the original claim unresolved.

• Was this standard of evidence stated (or clearly implied) before this round of evidence arrived, or is it being announced only now, right after the evidence came in?
• Can the person demanding more, right now, name some specific thing that would actually satisfy them — or does every hypothetical answer also get waved away in advance?
• Is the new, stricter standard justified by something independent of the fact that the old one was just met (a flaw discovered in the old standard, genuinely new information), or does its only apparent function seem to be excluding the evidence that just arrived?
• Tracked across the whole exchange, has the standard moved in one direction only — always harder, never easier — regardless of what gets presented?

Legitimate when: Raising or tightening a standard is legitimate when the change is triggered by something other than the bare fact that the old standard was met — e.g. the original standard turns out to have been genuinely too weak (satisfied by evidence that later proves flawed, fabricated, or coincidental), new and relevant information becomes available that nobody could have anticipated, or the standard was explicitly provisional and flagged as open to revision from the outset. The test: would this same tightened standard have been announced even if the evidence had never been produced?

Built from: Wikipedia — Moving the goalposts · Logically Fallacious — Moving the Goalposts · Wikipedia — List of fallacies

Multiple Comparisons Fallacy

Trying twenty different things, finding the one that looks impressive by luck, and reporting only that one.

Rina exported a year of her sleep tracker data and spent a Sunday running it against everything else she logs — meals, workouts, caffeine, screen time, weather, about thirty things in all. Oatmeal came out looking great: on oatmeal nights she slept twenty-two minutes longer, and the little significance badge lit up green. By Wednesday she had told four people that oatmeal is a sleep aid. What she hadn't mentioned to any of them were the other twenty-nine things she checked, one of which was always going to look like that.

The tell: The finding is oddly specific and arrives without a reason attached — nobody predicted it, it just turned up. If you ask what else they looked at and get a shrug, or a slightly defensive "well, a few things", you are looking at the winner of a contest you weren't told was running.

The move: A body of data is subjected to k separate significance tests, each at a threshold alpha. If nothing is going on, each test has probability alpha of returning a false positive, so the probability that at least one of them does is 1 - (1 - alpha)^k — at alpha = 0.05 and k = 100, about 0.994. The arguer reports the comparison or comparisons that cleared the threshold and omits, or simply never mentions, the count k. The surviving p-value is then read as though it had come from a single pre-specified test rather than from the maximum of many.

Why it fails: A p-value states how surprising a result would be if nothing were going on, and that statement is about the one comparison it was computed for and no other. Run enough comparisons and surprising results stop being surprising; they become the expected yield of the procedure. What governs the credibility of the reported finding is therefore not alpha but the family-wise error rate across all k tests, and k is precisely the number the report leaves out. There is a second defect on top of the first: because the winner is chosen after the outcomes are seen, it is selected for having the most extreme value, so its estimated effect is biased upward by the act of selection and will shrink on replication. The same data has been used twice, once to find the hypothesis and once to test it. Nothing about the reported test is technically wrong; what is wrong is the denominator.

• How many comparisons, outcomes, subgroups or time windows were examined in total — and does the report say?
• Was this particular comparison specified before the data were seen, or picked out afterwards?
• Was any adjustment made for the number of tests — Bonferroni, false discovery rate, a pre-registered analysis plan?
• Has the finding been reproduced on data that played no part in generating the hypothesis?

Legitimate when: Testing many things is not the problem; hiding the count is. The same analysis is sound when the number of comparisons is reported and the threshold adjusted for it, when a false-discovery-rate procedure is used for openly exploratory work, or when the analysis was pre-registered so that the comparison reported is the one that was promised. It is also sound as hypothesis generation, explicitly labelled: 'this turned up in the data and we are now going to test it properly on a fresh sample' is exactly the right use of a fishing expedition.

Built from: Wikipedia — Multiple comparisons problem · Logically Fallacious — Multiple Comparisons Fallacy

Naturalistic Fallacy

also known as: is-ought fallacy

Treating "this is how nature does it" as if that settled the question of whether it's any good.

At the Saturday market Hollis is selling raw honey and tells a young couple that raw is simply better — "nothing added, nothing cooked, exactly the way the bees make it." The woman asks what raw does that the pasteurized jars don't, and Hollis says it again, a little slower: it's what nature makes. Nobody at the table mentions that raw honey is one of the few foods you are told never to give a baby under a year old. Being the way the bees make it turned out to be the entire argument, and it isn't one.

The tell: You get a description dressed up as a verdict. Every time you ask "and that's better because—?" the answer is the same fact again, said with more feeling. Push once and you'll notice the speaker never has to say what the good thing actually is; the word 'natural' has been quietly carrying it the whole time.

The move: Two distinct moves travel under this name and both belong in the record. In G. E. Moore's original sense (Principia Ethica, 1903), the arguer identifies the property good with some natural or descriptive property N — pleasantness, being more highly evolved, being what we desire to desire — and then treats facts about N as settling evaluative questions, as though 'X is N' simply meant 'X is good'. In the looser popular sense, N is fixed as naturalness itself and the argument runs: X is natural, therefore X is good; X is unnatural, therefore X is bad. In both, an evaluative step is performed by a term with only descriptive content, and the bridge that would license the step is never stated.

Why it fails: Moore's test is the open question. If 'good' really meant N, then 'X is N, but is X good?' would be as closed as 'X is a bachelor, but is X unmarried?' — a question that answers itself the moment it is asked. It is not: for every candidate N ever proposed, the question stays substantive, and people who agree entirely about the facts concerning N go on disagreeing about the goodness. That shows the identification is not an analysis of what 'good' means but a substitution, with the evaluative work done by an unargued step hidden inside a definition. The popular form fails the same way with N fixed as naturalness: cyanide, cholera and infant mortality are natural, cochlear implants and antibiotics are not, so naturalness plainly does not track goodness and anyone using it owes an argument for the bridge rather than an assumption of it. Honesty requires one caveat the corpus should carry: many philosophers hold that Moore's target is a definitional and metaphysical error rather than a defective step of reasoning, and that calling it a fallacy at all is a misnomer — the Stanford Encyclopedia notes Moore himself denied he was analysing the meaning of a word.

• Is a word like 'good', 'better' or 'ought' being cashed out as a purely descriptive property — natural, common, older, pleasant, more evolved?
• Apply the open question: does 'yes, it has that property, but is it actually good?' still sound like a real question rather than a silly one?
• Is the bridge from the descriptive fact to the evaluation stated anywhere, or is it being carried by the choice of word?
• Do the obvious counter-cases survive — natural things that are terrible, artificial things that are fine — and does the speaker have a principled reason for setting them aside?

Legitimate when: Facts about nature are perfectly good evidence in an evaluative argument once an evaluative premise is on the table and defended. 'This pesticide is unnatural' does nothing on its own; 'this pesticide leaves residues that damage kidneys, and damaging kidneys is bad' does the work honestly, and its first half is a fact about nature. Appeals to the natural are likewise legitimate wherever 'natural' has been given a specific and relevant meaning — an unmodified control condition, an untreated crop, a food with no added sugar — and the evaluation rests on that specific property rather than on the honorific.

Ancient: Aristotle reasons from facts about the characteristic function of a human being to a conclusion about what a good human life consists in. Readers in the Moorean tradition treat that derivation as the standard ancient specimen of getting a value out of a natural fact. It is offered here as the discussion's usual reference point rather than as an attested error: many scholars deny the function argument is fallacious at all, and Moore's own targets were nineteenth-century naturalisms — hedonism, evolutionary ethics — not Aristotle. — Aristotle, Nicomachean Ethics I (the function argument)

Built from: Stanford Encyclopedia of Philosophy — Moore's Moral Philosophy · Wikipedia — Naturalistic fallacy · Wikipedia — Appeal to nature · The Fallacy Files — Appeal to Nature · Stanford Encyclopedia of Philosophy — Hume's Moral Philosophy

Negating Antecedent and Consequent

Flipping a rule inside out: hearing "if you do this, that happens" and taking it to mean "if you don't, it won't."

Dad's rule was stated once, in the driveway: "If you're not home by eleven, you lose the car for a week." Nolan gets in at 10:52, feeling clever, and by Saturday he has already told his friends he's got the car. When his dad takes the keys anyway — for the dishes, the two weeks of dishes — Nolan is genuinely outraged. "You said eleven!" What his dad gave him was one way to lose the car. Nolan heard it as the only way.

The tell: You feel like you were promised something that was never actually promised. Somewhere between hearing the rule and repeating it back to yourself, a warning about one route to the bad outcome turned into a guarantee that avoiding that route keeps you safe — and the guarantee is the part you can't find in anything anybody said.

The move: A conditional is granted: if P then Q. The arguer negates both halves in place and asserts the result as following: if not-P then not-Q. Both parts have been negated but not exchanged, which yields the inverse of the original rather than its contrapositive; the textbook name for the move is improper transposition. The same slip run backwards takes 'if not-P then not-Q' to 'if P then Q'. Note that what is drawn here is itself a conditional, which is what separates the move from denying the antecedent, where not-P is asserted outright and not-Q is drawn as a flat claim.

Why it fails: A conditional is equivalent to its contrapositive and to nothing else in the neighbourhood: 'if P then Q' says exactly what 'if not-Q then not-P' says, and it does not say what the converse or the inverse say. Proper transposition negates both halves and also swaps them; negating without swapping leaves the arrow pointing the wrong way through the negations. Put concretely, 'if P then Q' rules out one combination and one only — P together with not-Q — and it leaves not-P together with Q entirely open, which is precisely the combination the inverse claims to have ruled out. So a true statement yields a false one: if there is a fire there is smoke, but it does not follow that with no fire there is no smoke, since damp leaves and burnt toast make smoke without any fire. The pull of the move comes from hearing a sufficient condition as though it were the only condition.

• Was the original if-then merely rearranged, or were its two halves actually swapped as well as negated?
• Does the rule say the first thing is enough to get the second, or that it is the only way to get the second?
• Can you name a case with the second thing present and the first thing absent? If you can, the new if-then is broken.
• Would the speaker also accept the mirror version — second thing, therefore first thing — and if not, what makes this direction different?

Legitimate when: The inverse does follow when the original conditional is really a biconditional — when Q holds if and only if P does. Definitions, eligibility rules and switches behave this way: 'you qualify for the discount if and only if you are a member' licenses both 'member, so discount' and 'not a member, so no discount'. The honest form of the move is therefore to check whether the converse holds as well and to say so out loud, rather than to slide from a one-way rule to a two-way one in silence.

Built from: Wikipedia — Denying the antecedent · The Fallacy Files — Improper Transposition · Wikipedia — Contraposition

Negative Conclusion from Affirmative Premises

Using two statements about what things are to prove something about what things aren't.

At the food bank's Saturday briefing, Ines is going down the volunteer roster with a highlighter. "Some of our regulars are retired nurses," she says, "and every nurse has a current CPR card." Then she starts a second column. "So some of the regulars don't have one — we'll need to run a class for them." Bev looks up from the crates. "Which ones did you just find out don't have it?" Ines opens her mouth and closes it again, because nothing she said told her that about anybody.

The tell: Two cheerful facts go in and a shortage comes out. If you try to point at the actual people or things the shortage applies to, nobody can name one — the group of have-nots was produced by the sentence rather than found anywhere in the room.

The move: In the traditional analysis a categorical statement is affirmative (A: all S are P; I: some S are P) or negative (E: no S are P; O: some S are not P). This move starts from two affirmative statements and draws a negative one — for instance 'all colonels are officers; all officers are soldiers; therefore no colonels are soldiers', or, in the shape people actually commit, 'some volunteers are nurses; all nurses are certified; therefore some volunteers are not certified'. The rule broken is stated in traditional logic as: if both starting statements are affirmative, what follows from them must be affirmative too.

Why it fails: Affirmative statements assert inclusion or overlap and nothing else; a negative statement asserts exclusion, and no quantity of asserted overlap adds up to a single exclusion. The formal machinery makes this exact. A negative statement always distributes its predicate term — it says something about every member of that class — while an affirmative statement never distributes its predicate term. So if what is drawn is negative, its predicate term is distributed there, and if both starting statements are affirmative, that same term was distributed in neither of them. The argument therefore says more about that class than it was ever given, which is an illicit process of a term. The intuitive version is simpler: any diagram on which two affirmative statements come out true can be redrawn with everything nested inside everything else, so that nothing at all is excluded from anything.

• Are both starting statements affirmative — 'all', 'every', 'some are' — with no 'no', 'not' or 'none' among them?
• Does the thing being drawn contain a 'no', 'not', 'none', or a 'some are not'?
• Can you draw overlapping circles on which both starting statements are true and nothing whatever is excluded from anything?
• Is there an unstated negative doing the real work — something like 'and nobody else counts' — that was never actually said out loud?

Legitimate when: The traditional rule has a matching half: exactly one of the two starting statements must be negative if what follows is to be negative. So 'all nurses are certified; some volunteers are not certified; therefore some volunteers are not nurses' is fine, because a negative went in. And two affirmatives are perfectly capable of yielding a strong result — they simply yield an affirmative one, such as 'all colonels are soldiers'. The check is never whether the reasoning is ambitious, only whether an exclusion is being taken out that was never put in.

Built from: Wikipedia — Negative conclusion from affirmative premises · Wikipedia — Syllogistic fallacy

Nirvana Fallacy

also known as: perfect solution fallacy

Throwing out something that would help, because it wouldn't fix everything.

Jae tapes a chore chart to the fridge — four names, four columns, a Sunday reset. Marcus barely looks at it. "Chore charts don't work," he says. "Somebody always slacks off by week three and then we're right back here." He isn't wrong that somebody will slack off by week three. But the kitchen at this moment is at the stage where you can smell it from the hallway, and "this won't make us perfect" has just been used as an argument for leaving it exactly the way it is.

The tell: The objection is aimed at something nobody claimed. You offered a step and got told it isn't the destination. Notice too that the alternative on offer is never named — when you ask what we should do instead, the answer comes back as one more description of how the step falls short.

The move: A real option R is on the table and is compared not with the alternatives actually available but with an idealized alternative I that nobody has shown to be attainable. R falls short of I, and from that shortfall the arguer draws that R should be rejected, that R is worthless, or that there is no point adopting it. Harold Demsetz, who named the move in 1969, framed the defect as comparing an ideal norm against existing institutional arrangements instead of comparing alternative real arrangements with one another. The perfect-solution variant runs: R does not eliminate the problem entirely, therefore R is no solution at all.

Why it fails: A choice is always among the options that exist, so the relevant comparison for R is R against the other things that could actually be done — including doing nothing. Measuring R against an unattainable I answers a question nobody is in a position to act on. Because I is not available, rejecting R on I's account does not deliver I; it delivers whatever the status quo happens to be, and the status quo is itself an option that has quietly escaped the same scrutiny. The move also misreads what most measures are for, treating partial improvement as failure: a seatbelt never promised that nobody would die, and a hand-hygiene protocol never promised that nobody would be infected. Applied consistently the standard eliminates every course of action, inaction included, which is the surest sign it is not a standard at all.

• Is the option being compared against something someone could actually do, or against a described ideal with no route to it?
• If this option is rejected, what happens instead — and has that default been held to the same standard?
• Is the complaint that the option fails, or that it does not fully eliminate a problem it was only ever meant to reduce?
• Would the objection survive being restated as 'we should do this other available thing instead', and can the speaker name that thing?

Legitimate when: Comparing a proposal to an ideal is entirely proper when the ideal is a target for improving the proposal rather than a reason to bin it, and it is proper as a rejection when the shortfall is itself decisive — when a partial measure would do real harm, consume the budget for a better available measure, or create a false sense of safety that changes behaviour for the worse. It is also legitimate to reject a proposal by naming a better alternative genuinely on the table; that is the ordinary business of choosing, and it differs from this move precisely in that the alternative is real.

Built from: Wikipedia — Nirvana fallacy · Logically Fallacious — Nirvana Fallacy

No True Scotsman

also known as: appeal to purity

When someone loses the argument about a whole group, they quietly shrink the group until the awkward example doesn't count anymore.

In the group chat, Marco says, 'Real foodies always try the street food when they travel.' Priya replies, 'I'm a foodie and I skipped the street stalls in Bangkok, my stomach can't handle it.' Marco types back, 'Yeah, well then you're not really a foodie, are you?' Two minutes ago Priya was one of them. Now she isn't. The only thing that changed is that she disagreed.

The tell: the sudden goalpost move. A second ago the person was clearly inside the group, no questions asked. Now they're suddenly outside it, and the only new fact is that they said something inconvenient. It feels like the rule got rewritten live, just for them.

The move: 1. A universal claim is asserted: all members of group G have property P. 2. A genuine counterexample x is produced: x belongs to G but lacks P. 3. Instead of retracting or qualifying the claim, the arguer redefines G ad hoc as 'true G,' stipulating that x does not really belong, so the universal claim survives unrevised and without independent justification for the new boundary.

Why it fails: The redefinition is not driven by any independent criterion of group membership; it exists only to exclude the counterexample that just appeared. This turns an empirical, checkable generalization into a disguised tautology — 'every true G has P' now means only 'every G-who-has-P has P,' which is true by stipulation and says nothing about the world. The arguer keeps the rhetorical force of the original claim while quietly discarding its content, and does so without ever presenting evidence that the counterexample was misclassified in the first place.

• Was the 'true G' criterion stated, in that exact form, before this counterexample came up — or does it exist only because this case needed excluding?
• If the new criterion were applied retroactively and consistently, would it disqualify other members of G who were accepted without objection before now?
• Is there an independent test for 'true G' that doesn't just restate 'has property P' in different words?
• Does the arguer offer any evidence that the counterexample is not really a member of G, or only assert the exclusion?

Legitimate when: Tightening a category is legitimate when the refined criterion is stated independently of the counterexample, is applied consistently to past and future cases alike, and tracks a real, checkable distinction (for example, a professional body's published certification standard) rather than merely re-describing 'has the disputed property.'

Built from: Wikipedia · Internet Encyclopedia of Philosophy - Fallacies · The Fallacy Files - No-True-Scotsman

Non Sequitur

also known as: non sequitur, it does not follow

An answer that's perfectly true and has nothing to do with the question — the Latin just means "it does not follow."

Nadia sends back a burger that is still cold and pink in the middle. The manager comes over, wiping his hands on his apron, and says: "Ma'am, we have been in this neighbourhood thirty-one years." It is true. It is genuinely something to be proud of. Nadia sits there for a second trying to work out how to disagree with it, and can't, and also still has a raw burger sitting in front of her.

The tell: You get a reply you can't argue with and don't feel answered by. The urge is to say "okay, but—" and then discover you have no idea where to put the 'but', because nothing said was wrong; it simply never touched the thing you asked about. That gap between "I can't dispute that" and "that settled nothing" is the whole sensation.

The move: Non sequitur is Latin for 'it does not follow', and it names a genus rather than a single pattern: an arguer offers P1 through Pn in support of C, and C simply does not follow from them. In its strict sense it covers every argument whose form fails, so every named formal error is a species of it. In practice the term is reserved for two things — formally broken arguments that have no more specific name of their own, and the colloquial case in which the offered support is not merely too weak but materially irrelevant, the reasons and the claim being about different subjects. Any record of this label must say plainly that it is a parent category with many narrower children beneath it.

Why it fails: In the strict sense the failure is structural: the statements are arranged so that they could all be true while the thing drawn from them is false, and their truth therefore constrains it not at all. In the colloquial sense the failure is one of relevance: the offered reasons may be entirely true and entirely admirable and still bear no probative relation to what is at issue, so believing them moves you no distance at all toward the thing they were offered for. Either way the defect is a missing connection rather than a false statement, which is why non sequiturs so often survive fact-checking untouched — every ingredient is accurate and the dish is still not food. The label needs handling with care because it is diagnostically empty on its own: calling something a non sequitur says that support is missing without saying where, and the useful work is always done by the narrower name.

• Can you construct a case in which everything offered as support is true and the thing being claimed is still false?
• Do the reasons and the claim share a subject matter at all, or has the topic changed somewhere between them?
• Is there a more specific name for what went wrong — reversing an if-then, changing the subject, generalizing from one case? If so, use that name instead.
• If you granted every reason offered without argument, would you be even slightly closer to the thing being claimed?

Legitimate when: This is the one label with no legitimate twin, because it just is the name for support failing; the real risk runs the other way, in the label being applied too fast. Arguments get called non sequiturs when the connecting step is present but unstated — an enthymeme, where a shared assumption is left out because everyone in the room already holds it — and when the support is genuinely probabilistic rather than airtight, which is no defect in an argument that never claimed more. Before using the label, supply the missing step charitably and see whether it is one the speaker would plainly accept; if it is, the argument is compressed, not broken.

Ancient: Aristotle treats what he calls the refutation depending upon the consequent, which he says 'arises because people suppose that the relation of consequence is convertible' — knowing that if A holds B must hold, they take it that if B holds A must hold. His examples are bile mistaken for honey because honey is attended by a yellow colour, and rain read off wet ground: 'since after rain the ground is wet in consequence, we suppose that if the ground is wet, it has been raining; whereas that does not necessarily follow.' The Latin tag non sequitur is a much later coinage, but the diagnosis of something drawn that does not follow is his. — Aristotle, On Sophistical Refutations 5

Built from: Wikipedia — Non sequitur (logic) · Internet Encyclopedia of Philosophy — Fallacies · Aristotle, On Sophistical Refutations (Pickard-Cambridge trans., MIT Internet Classics Archive)

Notable Effort

Believing something because of how hard the person worked on it, rather than because of what they actually showed.

At the county science fair a judge stops at Table 19, where a ninth-grader named Priyanka has built a three-panel display on whether music helps plants grow: forty-one days of photographs, hand-drawn charts, a logbook with the date at the top of every page. The plants were on different windowsills and nobody measured the light. When the judges confer, one of them says he can't bring himself to mark her down — "look at that logbook, she did more work than anyone in this room." She did. The windowsills still wreck the result, and the logbook can't fix them.

The tell: You find yourself agreeing out of respect. There's a small pressure in the chest that says objecting now would be mean, and you notice you're weighing how much the person put in rather than what they put in front of you. That squeeze — "they tried so hard" arriving exactly where "and here's why" should be — is the thing.

The move: An arguer A advances a claim X and has visibly expended great effort in its service — years of study, a long document, a personal sacrifice, an evident sincerity of purpose. An audience takes that effort as support for X: A worked this hard on X, therefore X. In the sharper form the effort is offered by A directly, as a reason the case should be accepted or the objection dropped. The effort is a fact about A's conduct; the claim is about the world; and nothing has been said that connects the two.

Why it fails: How hard somebody worked on a claim is a fact about the worker and not about the claim. Effort does not track truth: a person can labour a decade on a false hypothesis and another can be right by lunchtime, and the world takes no notice of either. The defect is a relevance failure with an emotional engine — admiration and sympathy for visible work get converted into credence, and the conversion feels like fairness rather than like an error, which is what makes it hard to name aloud in the room. It is also socially expensive to resist, since whoever presses the objection appears to be belittling the work; the effort thereby acquires a second, protective function of making criticism look unkind. The decisive test is that the identical effort is available on both sides of any question, and effort on both sides cannot make both sides true.

• Is what's being offered a reason to believe the claim, or a description of what it cost to produce?
• If exactly the same quantity of work had gone into supporting the opposite claim, would it be equally convincing?
• Would the claim's support change at all if you learned the work had actually taken a single afternoon?
• Is anybody hesitating to press the objection because pressing it would seem to insult the effort, rather than because the objection has been answered?

Legitimate when: Effort is on topic in three ordinary situations. When what is being assessed is the person rather than the claim — a grade for diligence, a decision about whom to trust with the next job, the allocation of credit — effort is the very thing under evaluation. When the claim is itself a claim about search, effort is direct evidence for it: 'we checked all four hundred records' makes 'nothing was missed' more credible precisely because of the labour involved. And thoroughness is a genuine epistemic virtue, so a long inquiry can be better evidence than a short one when the length was spent covering cases rather than defending a position.

Built from: Logically Fallacious — Notable Effort · Wikipedia — Appeal to pity · Internet Encyclopedia of Philosophy — Fallacies

Nutpicking Fallacy

Digging up the worst three things anyone in a group ever said and holding them up as what the whole group is like.

Yusuf's sister has been on a knitting forum for two years and talks about it constantly. He finds the whole thing faintly ridiculous, so one evening he sends her three screenshots: someone threatening to report a stranger to a yarn company, someone calling a beginner's scarf "an insult to fibre", someone forty posts deep in an argument about whether machine-washing is a moral failing. "Your people are unhinged," he texts. There are eleven thousand people on that forum. He went looking, and he came back with three.

The tell: You recognize the quotes as real and the picture as wrong, and can't work out how both can be true at once. The giveaway is the sourcing: nobody named is anybody in particular, the examples arrive as screenshots rather than as anything you've seen yourself, and to say anything at all you'd have to defend strangers you would also roll your eyes at.

The move: A group G is at issue. The arguer selects instances from G's extreme tail — its most outrageous members, or the ugliest comments in an unmoderated thread — and presents them as showing what G thinks, wants, or is like. Schematically: fringe members of G assert or do X; therefore G asserts X or is X. The instances are genuine, which distinguishes the move from fabrication; the whole defect lies in how they were drawn. The term was coined on Kevin Drum's Washington Monthly blog on 11 August 2006, from a reader's suggestion, for the practice of trawling open comment threads to find a few extreme voices that can be held up as representative.

Why it fails: A generalization about a group is only as good as the sampling that produced it, and this sample was drawn by a procedure that explicitly hunts for the tail. Any group of appreciable size contains people who will say very nearly anything, so the existence of such people is close to certain in advance; finding them therefore tells you nothing you did not already know and nothing whatever about the distribution. The sample is not merely small — small samples are at least unbiased in expectation — it is selected on the very variable being measured, which is the worst defect an estimate can have. A second failure sits on top of the first: because the extreme position is far easier to refute than the group's typical or official one, defeating it produces the appearance of having answered the group while leaving its actual case entirely untouched.

• How were these examples found — by looking at what the group generally says, or by searching for the worst thing anybody in it said?
• Do the people quoted hold any office, authorship or standing in the group, or are they anonymous members of a crowd?
• Could you find comparably extreme voices attached to almost any large group, including one you belong to yourself?
• Has the group's stated or typical position been quoted anywhere in the argument, alongside the fringe one?

Legitimate when: The same evidence is entirely legitimate when the claim under test actually is about the tail: 'this movement has a violent fringe and does nothing about it' is properly supported by fringe examples together with evidence of the doing-nothing. It is legitimate when the people quoted are the group's leaders, official channels or most-followed voices, since then they are not the fringe but the group speaking. And it is legitimate when the sample was drawn without regard to extremity — a random pull of a hundred posts that happens to contain twenty ugly ones is a real measurement, and the number is the point.

Built from: Wikipedia — Straw man (weak man, hollow man, nutpicking) · Word Spy — nutpicking · Wikipedia — Faulty generalization

Overextended Outrage

Being rightly furious at a few people, then aiming the fury at everyone who resembles them.

An off-leash husky nipped a six-year-old at the Fremont dog park on Saturday, and by Sunday night the neighbourhood app has forty comments. Hal's is the one everybody shares: husky owners are the problem, they always have been, they should be banned from the park. There are maybe nine huskies that use that park. Nobody in the thread has said a word about the other eight, or about the owner who was on her phone by the gate, or about the two collies who have been snapping at people since June.

The tell: You agree completely with the anger and feel uneasy about where it's pointing. There's a specific move you can feel happen: a sentence starts with one person's name and finishes with a category. If you try to slow it down, you get treated as though you were defending what happened.

The move: Some members of a group G have done something genuinely outrageous, and outrage against those members is warranted. The arguer extends the indictment from them to G as a whole: some members of G did X; therefore G merits the anger owed to X, or G is the kind of group that does X. The instances are typically rare, are typically presented with no base rate alongside them, and the emotional charge does the work that a comparison of frequencies would otherwise have to do.

Why it fails: The step from 'some did' to 'they are like this' requires the instances to be representative, and rarity is exactly what makes them not. Outrageous acts become newsworthy in rough proportion to how uncommon they are, so the very salience that brought a case to attention is evidence against its typicality — availability runs inversely to frequency. Anger then supplies the missing warrant emotionally: it makes a case feel like a pattern, it makes the request for a denominator look callous, and it rewards whoever raises the temperature further. Two costs follow. The indictment lands mostly on members who did nothing, which is unjust on its face; and it obscures the actual wrongdoers by dissolving them into a class, so nothing is done about the thing that caused the anger in the first place.

• How many cases are actually being cited, and out of how many members of the group?
• Has any rate been given — is this more common in this group than in the comparison population, or is the comparison simply absent?
• Is the group being asked to answer for something it did as a group, such as a policy or a cover-up, or only for what individuals within it did?
• Would you accept the same reasoning about a group you belong to, on the same number of cases?

Legitimate when: Anger at a group is warranted when the group is what did the thing: when a policy authorized it, when the structure sheltered the people who did it, when complaints were made and buried, or when the rate really is elevated and the elevation has been measured rather than felt. It is also fair to hold a group responsible for what it declines to disown, since a leadership that will not act on repeated cases has made those cases institutional. The distinguishing question is always whether the argument runs through something the group did, or only through the arithmetic of a few of its members.

Built from: Logically Fallacious — Overextended Outrage · Wikipedia — Faulty generalization · Internet Encyclopedia of Philosophy — Fallacies

Oversimplified Cause Fallacy

Picking the one cause you can name and treating it as the whole story.

Sales at the Cardinal Street café have been down about a fifth since February, and Ivo is certain it's the rebrand. He paid real money for that logo, everyone said the old one was warmer, and by March he is reprinting cups, aprons, the awning, the lot. What doesn't make his list: the city tore up the block in January and the sidewalk is still a plywood ramp, he put a dollar on the drip in the same week, and Rosalind — the one everybody actually came in to talk to — took a job at the hospital. April comes back down another four percent.

The tell: The explanation is a little too tidy and it arrives with a plan already attached. Listen for the definite article: "the reason", "what happened was", "this is why". If you can name a second cause off the top of your head in five seconds and it wasn't on the list, then the list wasn't a list.

The move: An outcome Y is produced by several jointly contributing causes C1 through Cn. The arguer picks out one of them, Cj, and treats it as the cause: Cj contributes to Y, therefore Cj explains Y, therefore removing Cj removes Y. The move can run at the level of assertion rather than intervention — 'the reason for Y is Cj' — with the remaining contributors not denied so much as never mentioned. What has happened either way is a promotion: a contributing factor has been silently upgraded into a sufficient and exclusive one.

Why it fails: Most outcomes of any interest are produced by causes that are individually insufficient and jointly sufficient, so the truth of 'Cj contributes to Y' is entirely compatible with the falsity of 'Cj is why Y' and with the falsity of 'no Cj, no Y'. The oversimplification therefore does not merely lose detail; it licenses a prediction the evidence does not support, namely that acting on Cj alone will move Y by the amount credited to it. It also hides where the leverage actually sits: if Cj accounts for a tenth of the effect and Ck for half, a program aimed wholly at Cj will read afterwards as a refutation of the whole theory rather than as a misallocation. And the simplification is usually motivated — the surviving cause tends to be the one cheapest to fix, easiest to blame, or most congenial to whoever is choosing, which is why it so rarely runs the other way.

• Name a second cause that plausibly contributes to this outcome. Was it mentioned, ruled out, or simply absent from the account?
• Is the claim that this factor contributes, or that it is the reason — and did the speaker slide from one to the other mid-sentence?
• If this factor were removed entirely, is there any evidence about how much of the outcome would remain?
• Is the single cause named the one that happens to be cheapest to fix or easiest to blame?

Legitimate when: Singling out one cause is legitimate and often necessary. It is fine when the other contributors have actually been measured and turn out to be small; when the named cause is the only one anybody can act on, so the practical recommendation is unaffected even though the explanation is partial; and when the speaker is openly offering a simplification for a decision rather than an account of the mechanism. The honest form keeps the quantifier visible — 'the largest single driver', 'contributes substantially', 'accounts for about a third' — instead of the bare definite article.

Built from: Logically Fallacious — Oversimplified Cause Fallacy · Wikipedia — Fallacy of the single cause · Wikipedia — Questionable cause

Overwhelming Exception

A big impressive promise, with the fine print quietly cutting it down to almost nothing.

The window at Ironside says CANCEL ANY TIME in letters a foot high, and Dana signs up in October. In March she tries to cancel. Any time turns out to mean: sixty days' written notice, delivered in person, between ten and four on a weekday, and not during the first twelve months of the agreement. Every one of those was in the contract she signed. Put together, they mean that on the day she read the window, "any time" meant one specific Tuesday the following autumn.

The tell: The claim shrinks while you're holding it. You agreed to something clear and now find yourself in a conversation where each condition is individually reasonable and the thing you agreed to has gone. Watch for "of course" and "obviously" arriving in front of the qualification — they're doing the work of making a huge carve-out sound like a formality.

The move: A generalization of the form 'all A are B' is asserted with the rhetorical force of a sweeping claim and is then qualified — in a subordinate clause, in small print, or in a trailing 'except of course' — by conditions C1 through Cn which between them remove most or all of the cases a hearer would have had in mind. The qualified claim that survives is defensible; the unqualified claim that was heard is not; and the speaker collects the credibility of the first while enjoying the persuasive force of the second. David Hackett Fischer names the move in Historians' Fallacies (1970) as a generalization that is accurate but comes with qualifications eliminating so many cases that what remains is far less impressive than what was advertised.

Why it fails: Two different claims are in play and the argument needs them to pass as one. The broad claim does the persuading — it is what the hearer takes away, repeats and acts on — while the narrow claim is what the speaker retreats to when challenged, and because the narrow claim is true the speaker never has to concede anything at all. That structure makes the assertion unfalsifiable in practice while leaving it fully misleading in effect. The defect is one of scope rather than of truth: no individual sentence in the package is false, so fact-checking passes it untouched, and what has actually gone wrong is that the quantifier the hearer heard and the quantifier the speaker will defend are two different quantifiers. The practical harm is that the exceptions usually cover exactly the cases the hearer cared about, which is not an accident but the design.

• Restate the claim with every qualification moved to the front of the sentence. Is it still worth saying?
• Roughly what share of the cases a listener would have in mind actually survives the exceptions — most of them, or hardly any?
• Were the qualifications delivered with the same emphasis as the headline claim, or in a lower register: faster, smaller, later, elsewhere?
• If challenged, would the speaker defend the sweeping version, or fall back to the narrow one while keeping the credit earned by the sweeping one?

Legitimate when: Real generalizations almost always carry exceptions, and stating them is a virtue rather than a defect. The move is legitimate when the qualifications leave the claim's practical force intact — 'these are dishwasher safe, except the wooden handles' still tells you what you needed to know — when they are stated with the same prominence as the claim itself, and when what remains is what the hearer was actually asking about. The test is never whether exceptions exist, only whether, once they are all laid in front of you, the sentence still says what it sounded like it said.

Built from: Wikipedia — Overwhelming exception · Wikipedia — Faulty generalization

Package-Deal Fallacy

Assuming that because two things usually come together, you can't have one without the other.

At Thursday practice Omar says he wants to cut the horns from the slow one — just bass, drums and voice for four minutes, then bring them back for the last chorus. Delroy doesn't let him finish. "We're a soul band. Soul bands have horns. You take the horns out, you're not playing soul any more, you're playing something else." Nobody in the room can name a rule that says so, and half the Marvin Gaye in Delroy's own car is quieter than what Omar just asked for. What Delroy has is a habit, held up as a law.

The tell: You said one thing and got answered for five. There's a moment where you can feel an entire position being handed to you that you never asked for, and the strange part is that arguing starts to feel like betraying a team you don't remember joining. Listen for "so you're saying" followed by something you didn't say.

The move: Two or more items X and Y are customarily found together — as positions within a tradition, as tastes, as features of a product, as parts of an identity. The arguer treats the customary grouping as a necessary one: X and Y go together; S holds or wants X; therefore S holds or wants Y, or therefore X cannot be had without Y. In the variant Ayn Rand attached to the name, the bundling happens inside a single word, which is made to refer to two things that differ in exactly the respect under discussion, so that whatever is established about the one transfers unnoticed to the other.

Why it fails: Customary association is a fact about how a bundle came to be assembled — by history, by marketing, by coalition-building — and not a fact about whether the items entail, require or support one another. Nothing about two things being sold together makes either a condition of the other, so the argument needs a premise it does not have: that the connection is more than habit. Where the bundling lives inside a word, the failure is an equivocation, the term retaining its familiar reference while the argument quietly leans on a different one, so that the audience's agreement with the first is collected on behalf of the second. The practical damage is that the move erases the position actually in the room: a person who accepts one item and rejects the other finds their view has no name available to it, and gets answered as though they had said something they did not say.

• Is there a stated reason the two things must go together, beyond the observation that they usually do?
• Can you describe somebody who holds one without the other — and does that person actually exist somewhere?
• Is a single word being made to cover both items, so that agreeing about one looks like agreeing about both?
• Did the person actually say the second thing, or was it supplied for them out of the bundle?

Legitimate when: Grouping is legitimate in three cases. When one item genuinely entails or requires the other — you cannot have the heated seats without the wiring loom, and you cannot want a jury trial without wanting jurors — the package is real. When the connection is offered as a probabilistic guess and marked as one ('you liked the first two, you'll probably like this'), nothing has been overclaimed. And when a package is a fact about what is on offer rather than about reasoning — a menu, a legislative bill, a subscription tier — the right response is to argue about the bundling itself, which is a real argument and not this move.

Built from: Wikipedia — Package-deal fallacy · Logically Fallacious — Package-Deal Fallacy

Poisoning the Well

Trashing someone before they have said a word, so that whatever they say next already sounds worthless.

At the neighborhood association meeting, Marcus leans over to the two people beside him just as Priya walks to the front with her folder. "Heads up — she's the one who sued the last board. Whatever she's about to pitch, she's got an angle." Priya spends ten minutes laying out repair estimates for the parking lot, with three written quotes attached. Nobody looks at the quotes. Afterward a neighbor shrugs and says, "Well, you know what she's after," and that settles it.

The tell: You notice you had already made up your mind about someone before they opened their mouth, and you cannot point to anything they actually said that you disagree with. If it is aimed at you, there is a trapped feeling: defending yourself somehow proves the point, and staying quiet proves it too. The room is not listening for whether you are right, it is listening for the thing it was told to expect.

The move: Before an opponent S has offered any argument, an arguer A presents adverse information about S — character, motive, group membership, past conduct, true or false — and frames it as the lens through which whatever S says next must be read. When S then advances claim P with reasons R, the audience discounts R by reference to the pre-planted description of S rather than by examining R. The most closed variant is self-sealing: the poisoning is phrased so that any denial by S counts as confirmation of it ("that is exactly what someone like him would say").

Why it fails: Adverse facts about a speaker are at best evidence about that speaker's reliability as a witness; they are never evidence about the truth of an argument the speaker has not yet made. An argument's reasons stand or fall on their own content, and if they are checkable they can be checked without knowing who offered them. Poisoning the well does not even reach the ordinary ad hominem's threshold of engaging with something said — it operates on the audience in advance, replacing evaluation with a prepared disposition, so the argument is discounted before it exists. The self-sealing variant is worse than merely irrelevant: by construction no reply can count against it, which forecloses the possibility of any evidence entering the discussion at all. What presents itself as a warning about a source is in practice an instruction to stop listening.

• Is the discrediting information delivered before the target has stated their argument, so that there is nothing yet to answer?
• Grant the adverse claim about the person as entirely true — does any specific thing the target is about to say thereby become less likely to be correct?
• Is the framing built so that the target's denial or rebuttal would itself count as confirming the charge?
• Would the reasons the target offers be checkable by someone who had never heard of the target at all?

Legitimate when: Warning an audience in advance is legitimate when the warning concerns testimony rather than argument, and names a specific, documented conflict of interest or record of fabrication that bears on how much unverified word-of-mouth to accept — a court's disclosure that an expert witness is paid by one side, for instance. The line is that a genuine credibility disclosure invites closer scrutiny of what follows and can be defeated by checking it, while poisoning the well tells the audience the checking is unnecessary.

Built from: Wikipedia — Poisoning the well · The Fallacy Files — Poisoning the Well · Logically Fallacious — Poisoning the Well · Internet Encyclopedia of Philosophy — Fallacies

Political Correctness Fallacy

Deciding whether something is true by how comfortable or uncomfortable it is to say out loud.

The book club is arguing about a novel's ending. Renée has three pages of notes on why the last chapter contradicts the setup, with page numbers. Before she gets past the second note, Gary waves a hand: "Come on, that's just the approved take — everybody in these groups says that about this book." Her notes point to specific paragraphs; nobody opens the book to check a single one. The group moves on satisfied that the question has been answered.

The tell: Your evidence never gets touched. What gets discussed instead is what kind of person says the thing you said, or how it lands socially — and you notice the same move would work just as well against the opposite claim. There is a flatness to it: whatever you bring gets sorted into a category and filed.

The move: A claim P is put forward with reasons R. Rather than engaging R, the arguer classifies P by its social or political palatability — P is "the approved thing to say," or P is "politically incorrect" — and treats that classification as settling whether P is true. The move runs in both directions with equal confidence: "P is politically incorrect, therefore P is false" and "P is what decent people say, therefore P is true." A third use in circulation treats a change of terminology as though it were a change in the underlying facts.

Why it fails: How comfortable a claim is to say in a given room is a fact about the room, not about the world the claim describes. Palatability tracks social cost, and social cost is set by audiences, norms, and eras — none of which are inputs to whether the described state of affairs obtains. The move is doubly unreliable because it is symmetric: the same reasoning that lets one speaker reject a claim for sounding fashionable lets another accept it for exactly that reason, and a test that delivers opposite verdicts depending on who applies it is measuring the applier rather than the subject matter. It also short-circuits the one thing that could settle the matter, since R is never examined. Where the move instead treats a renaming as a change in the thing named, the defect shifts kind: a label is confused with what the label picks out.

• Has anyone engaged the actual reasons offered for the claim, or only its social reception?
• If the same claim carried the opposite political flavour, would this speaker's verdict on it change?
• Is "everyone knows you can't say that anymore" being offered in the position where evidence should go?
• Is a change of wording being treated as though it had changed the underlying facts?

Legitimate when: It is entirely legitimate to argue that a norm of speech has real costs — that a topic is under-researched because it is uncomfortable, or that a euphemism obscures something people need to see. That is an argument about the effects of a norm, supported by evidence about those effects, and it leaves the truth of the underlying claim to be settled separately. The defect appears only when palatability itself is offered as the reason the claim is true or false.

Built from: Logically Fallacious — Political Correctness Fallacy · Wikipedia — Political correctness · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies

Post Hoc

also known as: post hoc ergo propter hoc, after this therefore because of this, post hoc fallacy

Just because one thing happened right after another doesn't mean the first thing made the second thing happen.

Your coworker starts wearing a new pair of "lucky" socks on Monday, and the office wifi -- which had been flaky for weeks -- happens to run smoothly all day. By Wednesday he's telling everyone the socks fixed the wifi, and he's not wearing any other pair until further notice.

The tell: It feels like a story more than a real explanation -- there's a neat little sequence of events but nobody has said *how* the first thing could possibly bring about the second. If you ask "okay, but what's the connection?" and the only answer is "well, it happened right after," that's the tell.

The move: A occurred, then B occurred; therefore A caused B. (Temporal succession alone is treated as sufficient evidence of a causal relationship.)

Why it fails: Temporal priority is a necessary feature of many causal relationships but is never sufficient on its own to establish one. The inference ignores coincidence, common causes (confounders that produce both A and B independently), reverse or bidirectional causation, regression to the mean, and the simple fact that countless events precede countless other events without causing them. Establishing causation requires ruling out these alternatives -- through mechanism, controlled comparison, or repeated correlation -- not merely noting that one thing came before another.

• Could B have happened even if A never had -- is there a plausible account of B with A removed?
• Is there a third factor (a confounder) that could have produced both A and B independently?
• Does the effect track the supposed cause across repeated occurrences, or is this a single sequence being generalized from?
• Is there a specified mechanism connecting A to B, or only the bare fact that A came first?

Legitimate when: When temporal succession is paired with independent evidence for a causal link -- a known mechanism, a controlled or comparative design (e.g., the effect appears with the cause and disappears without it), ruled-out confounders, or a consistent dose-response/repeated pattern. Succession is then one legitimate piece of a larger causal case rather than the whole of it.

Ancient: Aristotle lists, among the merely apparent lines of argument speakers use, taking what is not the cause for the cause -- treating something as the cause of an event simply because it happened at the same time as, or just before, that event. His example: the orator Demades claimed that Demosthenes' policy was the cause of every disaster that followed, on no stronger ground than that war came after it. — Aristotle, Rhetoric II.24.8

Built from: Wikipedia: Post hoc ergo propter hoc · Stanford Encyclopedia of Philosophy: Fallacies · Aristotle, Rhetoric, Book 2, Chapter 24 (Perseus Digital Library)

Post-Designation

Going through a pile of numbers until something odd jumps out, then announcing you found what you were looking for.

Mr. Okafor has his seventh graders log everything about the twenty tomato plants on the windowsill — soil brand, watering day, shelf position, cup colour, planting date, eleven columns in all. At harvest, one group notices that the four tallest plants all sat in blue cups. "Blue cups grow taller tomatoes," goes the poster title. With eleven columns and twenty plants, some column was always going to line up with the tallest four.

The tell: The finding sounds far too specific and too surprising for how casually it was discovered. Ask when they decided that was the thing to look for, and the answer is always "when we saw it in the numbers" — which is the whole problem. The result usually arrives with a story invented afterward explaining why it made sense all along.

The move: A body of data D is collected or observed first. Only afterwards is the hypothesis H — a pattern, a subgroup, a target — designated, and it is designated precisely because it is the thing that stands out in D. The fit between H and D is then offered as confirmation of H, as though H had been specified in advance and D had been an independent test of it. The number of alternative patterns that were scanned and passed over is not reported, and often not counted.

Why it fails: The evidential force of a fit between hypothesis and data comes from the risk the hypothesis ran: a prediction made in advance could have failed and did not. A hypothesis read off the data afterwards ran no such risk — it was selected because it fits, so the fit was guaranteed before anyone looked. Any sufficiently rich body of data contains a great many patterns a searcher is free to designate, which makes the probability of finding some striking-looking pattern approach certainty even when nothing but noise is present. The reported significance is therefore computed against the wrong reference class: it treats one comparison as though it were the only comparison, when in fact many were silently run and one survived. Nothing has been confirmed except that the searcher was willing to keep looking.

• Was this specific pattern written down as the thing being tested before anyone examined these data?
• How many other patterns in the same data could equally well have been picked out and reported as the finding?
• Was the number of comparisons actually searched disclosed, and was the reported significance corrected for it?
• Has the pattern been retested on a fresh, independent body of data that played no part in generating it?

Legitimate when: Exploratory analysis of existing data is a legitimate and necessary way to generate hypotheses; the defect is not in looking but in reporting what was found as though it had been predicted. The same shape is fine when the post-hoc origin is disclosed and the claim is downgraded to a hypothesis worth testing, when the pattern is confirmed on an independent sample, or when the significance is corrected for the number of comparisons actually performed.

Built from: Logically Fallacious — Post-Designation · Wikipedia — Texas sharpshooter fallacy · Wikipedia — Data dredging · Wikipedia — Multiple comparisons problem · The Fallacy Files — The Texas Sharpshooter Fallacy

Pragmatic Fallacy

Treating "it worked for me" as proof that a thing works and that the story behind it is right.

Dev's shoulder has been aching for three weeks. His cousin gives him a copper bracelet, he wears it, and by the following weekend the ache is gone. Now he brings a spare bracelet to every family gathering. "I don't care what anybody says — I felt it. The copper pulls the inflammation out." Nobody in the kitchen mentions that most three-week shoulder aches go away in about three weeks, bracelet or no bracelet.

The tell: The whole case is a story about how someone felt, told with real conviction, and the timeline is doing all the work. If you ask what would have happened without the thing, you get a look — as though the question were rude rather than the obvious next one. And the story almost always arrives with a mechanism attached that sounds scientific and was never tested.

The move: A practice, product, or doctrine X is adopted; an outcome the user values — relief, success, a feeling of clarity — follows. From "X worked for me, or for people I know," the arguer concludes both that X is efficacious and that the theory T offered to explain how X works is true. Two substitutions are made at once: an uncontrolled personal outcome stands in for a measured effect, and the apparent efficacy of a practice stands in for the truth of the explanation attached to it.

Why it fails: Someone who improves after doing something has established that the improvement and the doing coincided, not that the doing produced the improvement. Conditions fluctuate and most acute complaints resolve on their own; people typically start a treatment when they feel worst, so any later measurement drifts back toward their average no matter what was done; expectation and attention generate real reported improvement all by themselves. Only a comparison — the same people, or comparable people, without X — can separate the effect of X from the effects of time, expectation, and selection, and a testimonial supplies no comparison at all. Even where a genuine effect is established, it underdetermines the explanation: a practice can work for reasons entirely unlike the story its practitioners tell, so its success cannot be read back as confirmation of that story. The word "works" also slides between "I am satisfied" and "it produces the effect claimed," and the argument needs the first to be granted and the second to be inherited.

• Does "it works" here mean a measured effect against a comparison group, or does it mean the person felt better?
• What would have happened to the same person over the same period without X — has anyone checked, or is there no comparison at all?
• Were the people who tried X and got nothing counted, or does the evidence consist only of those who stayed to report?
• Even granting that X helps, does anything in the evidence favour the offered explanation of why it helps over a rival one?

Legitimate when: Personal outcomes are perfectly good evidence when the effect is large, immediate, reversible, and repeatable on demand — turning the key starts the car, and no trial is needed to know it. Uncontrolled experience is also legitimate as the source of a hypothesis worth testing, and as grounds for a personal decision under uncertainty where the cost of being wrong is small. The defect appears only when it is offered as grounds for a general claim about efficacy, or as confirmation of the theory behind the practice.

Built from: The Skeptic's Dictionary — pragmatic fallacy · Logically Fallacious — Pragmatic Fallacy · Wikipedia — Regression toward the mean · Wikipedia — Anecdotal evidence

Prejudicial Language

Picking words that already deliver the verdict, so that accepting the description means accepting the judgment.

The wedding-party group chat is deciding on a venue. Tasha writes: "Anyone who actually cares about Mira's day will back the lakeside place — I'm not going to sit here and defend cramming her into that sad little hall to save a few bucks." Nobody had mentioned saving money, and the other option is a restored 1920s ballroom that happens to cost less. Three people who preferred it go quiet, because the only way to argue now is to first argue that they care about Mira.

The tell: You find yourself needing to defend your character before you can get to your point. The description of your position, in their words, is one you would never sign — and if you object to the wording it sounds like you are dodging the substance. Notice the pull to just agree, so you are not the person who apparently does not care.

The move: A situation that could be described neutrally is re-described in terms carrying a strong evaluative charge — "scheme" for proposal, "raid" for reallocation, "surrender" for compromise — and the charged description is presented as though it were a finding. The verdict is then drawn from the charge rather than from anything the charge was earned by. A common sub-form attaches the evaluation to the audience's self-image instead of to the object: only serious people, only good parents, only anyone who truly cares believes P, so believing P becomes the price of remaining one of those people.

Why it fails: Loaded wording asserts an evaluation while presenting it grammatically as a description, which lets it slip past the scrutiny an openly asserted evaluation would receive. Nothing about a proposal changes when "reallocate" is swapped for "raid": the facts are constant across the substitution, so any difference in the audience's verdict must be produced by the word rather than by the situation. That means the charged term is doing no evidential work at all — it is doing affective work in the place where evidence is expected, and the audience mistakes the second for the first because both arrive inside the same sentence. The self-image sub-form adds a second defect: it converts disagreement into a confession of bad character, so what the audience says is governed by the cost of dissent rather than by the merits.

• Rewrite the passage swapping every charged word for the flattest available synonym — does any argument remain?
• Is the evaluative word asserting a verdict that has not been separately established anywhere in the passage?
• Would someone who disagrees with the conclusion accept the description as a fair report of the facts?
• Does the wording make disagreement a statement about the disagreeing person's character rather than about the issue?

Legitimate when: Charged language is legitimate when the charge has already been earned by evidence in the same passage, or is uncontroversial between the parties — calling a documented, adjudicated embezzlement "theft" is accurate description, not prejudicial framing. The test is order and defeasibility: an earned evaluation follows its support and can be withdrawn if that support fails, while prejudicial language arrives instead of support and is not offered as anything that could be checked.

Ancient: Aristotle observes that among near-synonyms a speaker may always choose the term that dignifies or degrades the same act, and that this is a standard resource of persuasion. His examples are exact: to call the man who begs a man who prays, or the man who prays a beggar, since both are forms of asking; Iphicrates calling Callias a "mendicant priest" rather than a "torch-bearer"; actors styling themselves artists; pirates now preferring to be called purveyors. The action described is identical in each pair; only the word's charge differs. — Aristotle, Rhetoric III.2

Built from: Logically Fallacious — Prejudicial Language · Wikipedia — Loaded language · The Fallacy Files — Loaded Words · Aristotle, Rhetoric III.2 (Perseus Digital Library)

Proof by Intimidation

Winning by making people afraid to look stupid, instead of by showing your work.

In the Tuesday engineering sync, a new hire asks why the caching layer won't double-write during a failover. The lead architect sighs, takes off his glasses, and says, "Obviously it can't. This is standard stuff — it's in every distributed systems course." Two juniors write that down. Nobody asks again. Six weeks later, in production, it double-writes.

The tell: You have a specific, answerable question and you swallow it — not because it got answered but because asking it would cost something. Afterwards you notice you still do not understand the step, and neither does anyone else you ask quietly in the hallway. The room agreed; nobody was convinced.

The move: An arguer asserts P with a display — of confidence, credentials, volume, dense jargon, or visible irritation — and marks the unargued steps with phrases that price the asking of questions: "clearly," "it is well known that," "anyone who has read the literature knows," "I won't insult the room by walking through the obvious." No support for P is produced. Instead the cost of voicing doubt is raised until doubt is not voiced, and the resulting silence is read back as assent.

Why it fails: Confidence, status, and irritation are properties of a speaker's delivery, not evidence about the world, and a room's silence records what the room was willing to risk saying rather than what it found convincing. The move is self-concealing in a way ordinary irrelevance is not: it removes the very signal that would expose the gap, because the people most likely to spot a missing step are also the people most exposed to looking foolish for asking about it. That is why "obviously" so reliably marks the weakest joint in an argument — it is deployed exactly where a justification would be hardest to supply. And the technique is indifferent to truth: the identical delivery works whether P is true or false, so encountering it transmits no information about which. What the audience takes away is a memory of certainty rather than a reason.

• Strip out every confidence marker, credential, and expression of annoyance — what support for the claim is left?
• Is a step labelled "obvious," "well known," or "trivial" at precisely the point where filling it in would be hardest?
• Is the response to a question an answer to the question, or a comment on the person asking it?
• Did anyone in the room actually agree, or did they merely stop objecting?

Legitimate when: Legitimate expert communication compresses steps constantly — a paper says "it follows by the standard argument" because the standard argument genuinely exists and any reader can go find it. The condition is redeemability without penalty: if someone asks, the speaker can supply or point to the omitted step, and the asker pays no social cost for having asked. The defect is not the compression but the pricing of the question.

Built from: Wikipedia — Proof by intimidation · Logically Fallacious — Proof by Intimidation · Wikipedia — Argument from authority · Internet Encyclopedia of Philosophy — Fallacies

Proof Surrogate

Saying "studies show" or "everybody knows" instead of naming anything a person could actually go look at.

Kayla's uncle is explaining at Thanksgiving why he won't let his kids near the school tablets. "It's well documented at this point — screens before twelve rewire the whole attention system. The research is overwhelming. I'm not going to sit here and list it." Kayla asks which research. He waves his fork: "It's everywhere. Look it up." Nothing has been put on the table that anyone present could check, agree with, or argue against.

The tell: You want to look something up and there is nothing to look up. The sentence had the shape of a fact with a source behind it, but when you reach for the source your hand closes on air. Ask, and the work gets handed back to you — "go read about it" — which is the reverse of how showing someone evidence is supposed to go.

The move: A claim P is asserted alongside a marker M — "studies show," "it is well documented," "research has repeatedly demonstrated," "informed sources say," "the science is settled" — where M implies that supporting evidence exists without identifying any of it. The audience is invited to treat P as though the evidence M gestures at had actually been supplied. No source, study, dataset, or expert is ever named that could be located, read, or contested.

Why it fails: The existence of evidence is not itself evidence; it is a further claim, and here it is presupposed rather than argued for. Marker phrases work by borrowing the grammatical form of a citation while omitting everything that makes a citation useful — the identity of the source, its methods, its scope, and whether it says what the speaker says it says. Because nothing is named, nothing can be checked, so the ordinary corrective process is blocked at the outset rather than merely made difficult. The move is also perfectly available to a speaker with no evidence at all, which means encountering it tells the audience nothing about whether the evidence exists; a signal that costs nothing to send carries no information. Delete M from the sentence and what remains is the bare assertion of P, which is in fact all that was offered.

• Which specific study, source, expert, or dataset is being invoked — can it be named, and is a name forthcoming on request?
• Delete the marker phrase from the sentence: what supporting material is left standing?
• Would the same sentence be equally sayable, word for word, by someone who had no evidence whatsoever?
• Is "everyone knows," "it's well documented," or "the research is overwhelming" sitting in the position where a citation would go?

Legitimate when: Summary reference to a body of literature is ordinary and necessary — no speaker can recite every trial from memory, and "the evidence on this is strong" is fine when the speaker can identify that evidence on request, when the audience already shares the reference, or when the claim is uncontested background rather than the point at issue. The condition is redeemability: a legitimate summary is a promissory note the speaker can cash, and it becomes a proof surrogate when the note is offered in place of payment and never honoured.

Built from: Logically Fallacious — Proof Surrogate · Wikipedia — Weasel word · Wikipedia — Ipse dixit · Internet Encyclopedia of Philosophy — Fallacies

Proving Non-Existence

Making a claim, then demanding that anyone who doubts it prove the thing never happened anywhere, and counting their failure as a win.

In a neighbourhood forum thread about a missing package, Ron posts that the drivers on this route "have been taking stuff for months." Someone asks what he has. "Prove they haven't," he replies. "You can't tell me not one driver has ever pocketed anything on this route. You can't — which is exactly my point." Twelve replies later nobody has produced a single missing-package report, and the thread has settled into treating the theft ring as established.

The tell: You are suddenly the one doing the work, and the work is impossible. Notice the shape of what you have been asked for: not "show me one case where this didn't happen" but "show me it never happened, anywhere, ever." And notice that no matter what you produce there is always more unsearched space left, which means the demand can never be met and was never meant to be.

The move: A claimant asserts that X exists, or occurred, without producing evidence for it. Challenged, the claimant does not supply evidence but demands that the challenger establish the universal negative — that X exists nowhere, occurred at no time, in no case. The challenger cannot discharge a demand nobody could discharge, and that failure is then presented as support for X: "you can't prove it didn't happen, so it did."

Why it fails: The demand is rigged, and the rigging is the whole of the move. Establishing an unrestricted negative requires exhausting an unbounded search space, so failure to establish it is guaranteed in advance and therefore carries no information about whether X exists — a test that every possible world passes distinguishes none of them. The move also inverts the ordinary allocation of burden. The party who introduces a positive existence claim is the party who has undertaken to support it, and for a good structural reason: the positive claim can be settled by a single instance, while its denial cannot be settled by any finite number of failures to find one. Converting the challenger's inevitable silence into support runs absence of evidence backwards, treating a gap the claimant created as though the claimant had filled it. Nothing has been shown except that the world is large.

• Who introduced the existence claim, and has that party produced a single positive instance of it?
• Would discharging the demand require searching an unbounded space — every place, every time, every case?
• Is the challenger's inability to complete that search being counted as support for the claim, rather than as a fact about the demand?
• What observation would the claimant accept as telling against X — is there any, or is the claim built so that nothing could count against it?

Legitimate when: Negative existence claims are often perfectly establishable and are legitimately demanded when the search space is bounded and the thing would leave traces: there is no elephant in this kitchen, no arithmetic error in this ledger, no such gene in this sequenced genome. It is equally legitimate to argue from absence when evidence should have turned up and did not — a well-designed search that comes back empty is real evidence against existence. The defect arises only when the space is unbounded, or the thing is defined so as to leave no traces, and the guaranteed failure is then scored as a point.

Ancient: The Roman jurist Paul lays down that proof falls upon the one who asserts, not upon the one who denies — ei incumbit probatio qui dicit, non qui negat. The maxim survives in the sixth-century Digest and is the ancestor of the modern rule that a party introducing a positive claim carries the burden of establishing it. It is a rule of procedure rather than an analysis of a reasoning defect, but it states precisely the allocation this move reverses. — Paulus, in Justinian's Digest 22.3.2

Built from: Logically Fallacious — Proving Non-Existence · Wikipedia — Argument from ignorance · Wikipedia — Burden of proof (philosophy) · Wikipedia — Legal burden of proof

Pseudo-Logical Fallacies

Slapping an official-sounding label on someone's reasoning and treating the label itself as the answer.

In a Discord argument about a game patch, Theo posts three numbers straight from the patch notes showing a weapon's damage went down. Mira replies: "Classic recency bias — textbook," and links a list she found. Nobody, Mira included, can say what that name is supposed to mean here or which of Theo's three numbers is wrong. The thread treats the matter as closed anyway, because the reply sounded like it came with authority behind it.

The tell: You have been named at, not answered. Ask which part of what you said the name is supposed to describe and there is no answer — the label was the entire move. It stings in a particular way, because it borrows the manner of careful reasoning while doing the opposite of it.

The move: An arguer answers a claim P not by identifying a defect in the reasoning offered for P but by attaching a fallacy label L to it — frequently a coined, non-standard, or ad hoc label — and treating the attachment as sufficient to dismiss P. No failure condition for L is stated, no step in the argument actually on the table is identified as the one that fails, and the name does the work an analysis would have done. As a corpus heading, "pseudo-logical fallacies" is the bucket for such labels: names in circulation that lack a stable definition, an identifiable failure condition, or any standing in the literature.

Why it fails: A fallacy name is shorthand for a diagnosis, and shorthand is only as good as the diagnosis it abbreviates. When the label carries no stated failure condition, invoking it asserts that something went wrong without saying what, which leaves the other party nothing to answer and the audience nothing to check — the appearance of analysis with none of its content. Even a correctly applied label does not transfer to the claim: showing that one argument for P is bad leaves the truth of P entirely open, so dismissing P on the strength of a name is itself a misstep. And because anyone may coin a name, an inventory that admits every coinage stops discriminating; users then cite a heading as though it picked out a specific defect when it picks out nothing at all, which is exactly the hazard this record exists to mark.

• Does the label being invoked have a stated failure condition — can the person using it say what has to be true of a passage for the label to apply?
• Has any specific step in the argument on the table been identified as the one that fails, independently of the name?
• Does the label appear in a standard reference work or critical-thinking text, or only in listicles and comment threads?
• Is the claim itself being treated as refuted because one argument for it was labelled, rather than merely left unsupported?

Legitimate when: Naming a fallacy is legitimate and useful shorthand when the name has a settled definition, the speaker can state its failure condition, and the specific step at fault is pointed to — "that's affirming the consequent, and here is the line where it happens" is analysis compressed, not analysis replaced. Even then the correct verdict is that this argument fails, not that the claim is false.

Built from: Logically Fallacious — Pseudo-Logical Fallacies · Wikipedia — Argument from fallacy · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies

Psychogenetic Fallacy

Explaining why someone believes a thing, and treating that as having shown the thing is wrong.

Nadia tells her brother the family should stop lending money to their cousin, and lays out the four times it was not paid back, with dates. Her brother does not touch the dates. "You've been like this since Dad's business went under — you panic about money. This isn't about Kofi, it's about you." Maybe it is about her. The four dates are still four dates, and nobody looks at them.

The tell: The reply is about your history, not your point, and it is usually delivered gently enough that pushing back feels ungracious. You catch yourself defending your motives instead of your evidence. Notice that even if the whole story about you were true, your evidence would still be sitting there unexamined.

The move: S asserts P and offers reasons R. Rather than examining R, the arguer supplies a causal story about S's psychology — bias, upbringing, fear, wish-fulfilment, wounded pride, trauma — that would explain S's holding P without P's being true, and treats the availability of that story as showing P false or unwarranted. The origin of the belief is substituted for the standing of the claim.

Why it fails: Every belief anyone holds has a psychological history, true ones included, so an account of how a person came to hold P is compatible with P's being true, false, well-supported, or baseless — by itself it sorts nothing. Causal stories about belief formation are also cheap to produce and hard to refute: a plausible motive can be constructed for any position whatever, including the position of the person offering the story, which means the move generalises into an argument against everybody and therefore against nobody. Two further defects usually ride along. The origin story is typically asserted rather than established, so an unevidenced psychological claim is being used to displace an evidenced one. And the exercise quietly presupposes its own conclusion, since one only reaches for a debunking explanation after having decided the belief needs explaining away rather than answering. Throughout, the actual support for P sits untouched and checkable.

• Has the offered psychological story been established, or merely supposed because it would be convenient?
• If a person with none of the alleged motives said exactly the same thing, would this response still apply to it?
• Does anything in the reply engage the reasons given for the claim, as opposed to the reasons the person holds it?
• Could an equally plausible motive-story be told about the person making this reply — and would that story settle anything?

Legitimate when: Psychological explanation is legitimate when the question genuinely is why a person believes something rather than whether it is so — in intellectual history, in therapy, in weighing how much credence to give unverifiable testimony from an interested party. It is also a real defeater for testimony specifically: if the only support for P is that S says so, and S has a demonstrated motive to say so regardless of the facts, the weight of that testimony really does drop. The line is that this lowers confidence in an unchecked report; it never refutes a claim that arrived with checkable reasons attached.

Built from: Logically Fallacious — Psychogenetic Fallacy · Wikipedia — Genetic fallacy · Wikipedia — Bulverism · Wikipedia — Ad hominem

Quantifier-Shift Fallacy

also known as: quantifier shift

Going from "everyone has one" to "there's one that everyone has," which are not the same thing at all.

At the start of the season the volleyball coach tells the parents, "Every girl on this roster has a position she's strongest at." The team parent repeats it in the group chat as "Coach says there's a position they're all strongest at," and by Thursday two families are arguing about which one it is. The coach meant eleven different answers, one per player. The retelling turned it into a single answer for all eleven.

The tell: A sentence gets repeated back to you and it sounds like what you said, but something has quietly gotten bigger. Test it on a case you know cold: everybody has a birthday, but there isn't one birthday everybody has. If swapping "everybody has some" for "there's some that everybody" flips a sentence from obviously true to obviously false, that same swap just happened to you.

The move: From a statement of the form "for every x there is some y such that R(x,y)" (∀x∃y Rxy), the arguer moves to "there is some y such that for every x, R(x,y)" (∃y∀x Rxy). The two quantifiers are permuted, converting a claim that allows a different y for each x into a claim asserting a single y that serves them all. The reverse permutation — from ∃y∀x Rxy to ∀x∃y Rxy — is unobjectionable, so the defect is directional: the error lies in one crossing and not the other.

Why it fails: The order of quantifiers fixes what depends on what. In ∀x∃y Rxy the choice of y is made after x is given and may vary with it; in ∃y∀x Rxy a single y is fixed first and must then work for every x. The second is strictly stronger, and nothing whatever in the first supplies the extra content the second asserts, namely a common witness. A one-line counter-model settles it: every person has a mother, but there is no one woman who is everyone's mother; every natural number has a successor, but no single number succeeds them all. Since the starting statement is true and the derived one false in the same model, the step cannot be licensed by form, and no appeal to how natural the English sounds repairs it. Ordinary language hides the defect because "everyone has something they're afraid of" and "there's something everyone's afraid of" differ by a word order the ear glides straight over.

• Write both statements out with their quantifiers explicit — is the order of "every" and "some" the same in each, or has it flipped?
• Does the second statement require a single shared thing where the first allowed a different thing for each case?
• Try to describe a case in which each item has its own witness but no witness is shared — if you can describe one, the step fails.
• Is the step running from "there is one that covers all" to "each has one" (fine), or from "each has one" to "there is one that covers all" (not fine)?

Legitimate when: The permutation is entirely legitimate in the other direction: from "there is a single master key that opens every door" it does follow that "every door has some key that opens it." It is also legitimate in the flagged direction when a further fact independently establishes the common witness — a uniqueness result, a finite domain with a maximum, a closure property, or a stated convention. The rule of thumb: you may always weaken a shared witness into individual ones; you may never strengthen individual ones into a shared one without a separate reason.

Ancient: The Ethics opens by observing that every art and every inquiry aims at some good, and moves in the same breath to speaking of "the good" as that at which all things aim. Since J. L. Ackrill, and in Harry Gensler's teaching version, commentators have standardly cited the passage as an alleged illicit quantifier shift — from "each pursuit aims at some good" to "there is one good all pursuits aim at." The attribution is genuinely contested: defenders read Aristotle as fixing the sense of "the good" rather than arguing to a single end. Recorded here as a documented modern allegation about a real passage, not as a settled ancient example. — Aristotle, Nicomachean Ethics I.2 (1094a1–22) — an alleged and disputed instance

Built from: Wikipedia — Quantifier shift · Internet Encyclopedia of Philosophy — Fallacies · Stanford Encyclopedia of Philosophy — Aristotle's Ethics · Stanford Encyclopedia of Philosophy — Fallacies

Quantum Physics Fallacy

Sprinkling physics words over an unrelated claim so that it sounds as though science is backing you up.

At a weekend workshop above a yoga studio, the facilitator tells the room that "the observer changes what's observed — that's not spirituality, that's laboratory science." Therefore, she says, writing your goal on a card and picturing it each morning literally alters which outcome occurs. Ellie, three chairs over, has a nagging sense that "observer" in a lab does not mean a person hoping. She is right, but nobody in the room can say why, so it lands as settled science.

The tell: The physics arrives at exactly the moment the argument needs support, and then it leaves — nothing else in the talk depends on it. You notice you would have to be a physicist to object, which is the point. Try asking what the word means in the actual experiment, in one sentence; if nobody can say, the word was not carrying knowledge, it was carrying authority.

The move: A claim C having nothing to do with quantum mechanics is supported by invoking quantum vocabulary — observation, entanglement, superposition, uncertainty, energy, the field — either as a straight appeal to the prestige of physics ("quantum physics has proven that...") or by trading on a term's colloquial sense while borrowing the technical sense's authority. Neither the cited result nor its domain of application is stated, and no route from the physics to C is shown.

Why it fails: Two defects run together. First, borrowed prestige is irrelevant: that a difficult and enormously successful theory exists says nothing about a claim the theory does not address, and naming it functions as an appeal to an authority which has not in fact spoken. Second, the key terms equivocate. "Observer" in quantum mechanics denotes a physical measurement interaction, not a conscious mind forming an intention; "uncertainty" names a precise relation between conjugate variables, not open-mindedness; "entanglement" describes correlations between prepared systems that provably cannot transmit information, not sympathy between people. The argument needs the technical sense to supply authority and the colloquial sense to reach the conclusion, and no single reading of the words supports both steps. Scale compounds the problem: the phenomena cited are established in regimes where decoherence has not washed them out, and nothing licenses transporting them to bodies, minds, or markets.

• Which specific result is being cited, and at what scale and under what conditions is it actually established?
• Is the key term ("observer," "uncertainty," "energy," "entangled") used in its technical sense in one step and its everyday sense in the next?
• Replace the physics vocabulary with plain words stating exactly what the physics says — does any support for the conclusion survive the replacement?
• Is any route offered from the cited physics to this conclusion, or is the physics named and then abandoned?

Legitimate when: Quantum mechanics is legitimately invoked wherever the phenomena really are quantum — semiconductors, lasers, magnetic resonance imaging, chemistry, cryptography, computing — and it is legitimate to draw philosophical conclusions from it when the argument engages the actual formalism and the live interpretive debates about measurement, non-locality, and determinism, which are serious and unsettled. The conditions are that the physics is stated accurately, its domain respected, and the path from it to the conclusion shown rather than gestured at.

Built from: Logically Fallacious — Quantum Physics Fallacy · Wikipedia — Quantum mysticism · Stanford Encyclopedia of Philosophy — Philosophical Issues in Quantum Theory · Wikipedia — Argument from authority

Questionable Cause

also known as: cum hoc ergo propter hoc, correlation implies causation

Two things move together, so one of them gets blamed for the other with nothing else to go on.

The manager of a corner bakery notices that on days when the pastry case is fully stocked by seven, sales run about a fifth higher. She starts paying a baker to come in an hour early. Sales do not budge. It turned out the early-stocked days were the days her fastest baker worked — and that baker was also the one who chatted with the regulars and talked them into a second coffee. Stocking time and sales moved together because someone else was moving both.

The tell: The story is told as if the pattern speaks for itself, and there is a jump you can feel between "these go together" and "this one makes the other happen." Ask what else changes on those same days and you usually get a pause. It also tends to arrive with exactly one candidate on the table — nobody mentions the other three things that would explain it just as well.

The move: A and B are observed to covary — co-occurring, correlated across a population, or moving together over time. From that association alone the arguer concludes that A causes B, leaving the direction of the relation, the mechanism connecting them, and every rival explanation unexamined. This is the genus the Latin tag cum hoc ergo propter hoc names, "with this, therefore because of this," and it covers any move from bare association to a causal verdict.

Why it fails: Association is symmetric and causation is not, so the evidence on offer cannot by itself pick out which of the two things is doing the causing: every correlation between A and B is equally consistent with A causing B, with B causing A, with some third factor producing both, and with coincidence. Selection can manufacture the association with no causal link at all, whenever the way cases entered the data depends on both variables. Where the first measurement was taken at an extreme, regression toward the mean produces movement that a naive reading credits to whatever happened in between. Discriminating among these possibilities requires something the bare correlation does not contain — an intervention, a temporal ordering with confounders held fixed, a known mechanism, or a design that makes the rivals implausible. Without one of those the causal claim is not weakly supported but unsupported in the one respect that matters: nothing in the evidence favours it over its competitors.

• Could the arrow run the other way — could B be causing A rather than the reverse?
• Is there a plausible third factor that would produce both A and B, and has it been measured and held fixed?
• Is there any evidence beyond the covariation — an experiment, a natural experiment, a temporal ordering, a mechanism?
• Were the cases selected in a way that depends on both A and B, so that the association could be an artefact of who got counted?

Legitimate when: Inferring cause from association is legitimate when the design earns it: randomisation breaks the link between treatment and confounders, so a correlation that survives it is causal. More weakly but still genuinely, natural experiments, instrumental variables, well-specified causal models with the confounders measured, dose-response relationships, and a known mechanism can all license the step. Association is also perfectly good grounds for prediction without any causal claim at all — you may forecast B from A without asserting that A produces it.

Ancient: In his catalogue of merely apparent enthymemes, Aristotle lists taking as a cause what is not a cause — treating something that merely accompanied or preceded an event as though it had produced it. He treats the same move in On Sophistical Refutations under the heading of the non-cause taken as cause, where a step is blamed for a result it played no part in bringing about. These passages are the classical root of both this genus and its narrower species, post hoc. — Aristotle, Rhetoric II.24 and On Sophistical Refutations 5

Built from: Logically Fallacious — Questionable Cause · Wikipedia — Questionable cause · Wikipedia — Correlation does not imply causation · Internet Encyclopedia of Philosophy — Fallacies

Rationalization

Deciding first for reasons you would rather not say, then building a respectable-sounding case afterwards.

Marisol tells her sister she is not going to the reunion because of the drive, and the gas, and how she would have to board the dog, and anyway the hotel block sold out. Her sister offers to drive, split the gas, take the dog, and knows a place two miles out. Marisol produces a fifth reason within about ten seconds. She does not want to see their aunt, which she has not said, and which was the whole thing from the start.

The tell: Every objection you answer is replaced by a fresh one, roughly as fast as you can answer them, and each new one is a little smaller than the last. You get the sense you are arguing with the front of a building. From the inside it is the feeling of hearing yourself say something you do not entirely believe while it comes out fluent and well organised.

The move: A position C is already held, or a decision already made, on grounds M that the arguer will not or cannot state — self-interest, loyalty, fear, appetite, prior commitment. Reasons R are then assembled after the fact and presented as the grounds for C. R was not what produced C, and R was selected by the test "does this support C and sound presentable?" rather than by the test "is this true and does it bear on C?"

Why it fails: The defect is not that R is false — every item in R may be true — but that R was never load-bearing, so the argument misreports its own structure and invites the audience to test the wrong thing. Reasons chosen because they support a fixed position carry no evidential weight about that position, in the same way that polling only the people who already agree with you tells you nothing about the population: the filter, not the world, determined what showed up. The reliable symptom is that R is not defeasible in the arguer's own hands — refute every item and the position does not move, which demonstrates that it was never resting on them. A stack of independent-sounding minor reasons is the usual signature, since a genuine ground rarely needs four spares. And because the stated reasons are not the operative ones, the exchange cannot converge: an interlocutor who answers R is dismantling scaffolding rather than the wall.

• Were the stated reasons formed before the position was adopted, or produced afterwards, in response to being asked?
• If every stated reason were conceded to be wrong, would the arguer change the position — or reach for another reason?
• Are the reasons multiplying and shrinking, several small ones offered where one that carried the weight would do?
• Does the same person apply these reasons consistently in cases where they would lead somewhere unwelcome?

Legitimate when: Reconstructing one's reasons after the fact is normal and frequently honest: people often reach a sound judgement by pattern recognition and only afterwards articulate why, and a reason discovered late can be a perfectly real reason. What makes the difference is whether the stated reasons are held defeasibly — whether the arguer would actually move if they failed — and whether they are applied evenhandedly to cases that cut the other way. A late-articulated ground the arguer would abandon the position for is a genuine ground.

Built from: Logically Fallacious — Rationalization · Wikipedia — Rationalization (psychology) · Wikipedia — Motivated reasoning · Stanford Encyclopedia of Philosophy — Self-Deception

Red Herring

It's when someone answers a question you didn't ask, hoping everyone forgets the one you did.

In the family group chat, Mom asks Dad why he spent the vacation fund on a boat without telling anyone. Dad replies, 'You know, it's pretty telling that nobody in this family ever notices how many hours I put in at work.' Within minutes the whole chat is arguing about whether Dad works too hard, people are sending heart emojis, and the boat never comes up again.

The tell: The conversation has drifted somewhere louder and more emotional, and it feels vaguely related to what you were originally asking — but when you try to steer it back, you're suddenly the one who looks petty or cold for 'still going on about' your original question.

The move: The question at issue is P. The arguer introduces a different claim R that is related enough in tone or topic to seem connected to P but does not bear on whether P is true. The arguer then argues R at length — often persuasively and even correctly — and the audience, following the energy of the exchange, treats the resolution of R as though it had settled P, while P quietly goes unanswered.

Why it fails: The defect is pure irrelevance, not falsehood: R can be entirely true and R's defense can be entirely sound, and none of that gives any evidential weight for or against P, because there is no logical bridge — no entailment, no shared premise doing real work — connecting R's truth to P's truth. The Stanford Encyclopedia of Philosophy's discussion of fallacies of relevance and the Internet Encyclopedia of Philosophy both place this under the broader heading of ignoratio elenchi ('irrelevant conclusion' / 'missing the point'): the argument that gets made may be internally fine, it just isn't an argument about the thing in dispute.

• Name the exact question on the table in one sentence — is the new topic R actually a claim that, if true or false, changes the answer to that question, or does it just live in the same neighborhood emotionally?
• If R were fully and conclusively settled right now, would the original question P be one inch closer to answered — or exactly as open as it was before anyone spoke?
• Who benefits from the audience's attention moving off P and onto R — does the shift happen right at the point where P was getting hard to defend?
• Strip out everything said about R — does the speaker have anything left that actually addresses P, or was R the entire response?

Legitimate when: Raising a second issue is legitimate, not a red herring, when it is flagged honestly as a separate point ('that's a fair question, but first — a related issue matters here too') rather than substituted for an answer, or when R genuinely is a necessary precondition, scope qualifier, or piece of context that changes how P should be evaluated — for instance, pointing out that the premise of a question is itself false, or that a decision has a relevant cost the original framing ignored, can be a real contribution rather than a diversion, provided the original question still gets addressed afterward rather than being permanently abandoned.

Ancient: Aristotle diagnosed a family of fake refutations as arising from 'ignorance of refutation' (ignoratio elenchi): to truly refute a claim you must disprove that exact claim, so proving or attacking something merely adjacent to it only has the appearance of a refutation. He treated this failure as underlying a great many of the sophistical tricks he catalogued. The nineteenth-century English term 'red herring' names the everyday, informal version of that same ancient move — swapping in a related-sounding topic for the one actually in dispute — though the fishy name itself postdates Aristotle by roughly two thousand years. — Aristotle, Sophistical Refutations (Sophistici Elenchi)

Built from: Wikipedia — Red herring · Wikipedia — Irrelevant conclusion · Internet Encyclopedia of Philosophy — Fallacies · Stanford Encyclopedia of Philosophy — Fallacies

Reductio ad Absurdum

also known as: reductio ad absurdum, reduction to absurdity, proof by contradiction

Testing an idea by taking it completely seriously and following it out until it runs into something nobody could accept.

Jonah's roommate insists nobody may touch the thermostat because "whoever's coldest decides." Jonah doesn't call the rule silly; he runs it. "Okay — so in July Priya's cold in the AC and sets it to eighty-two. Now I'm the coldest at eighty-two, so I set it to eighty-six. Then you're the coldest, and up it goes, and at every step the rule is being followed exactly." His roommate laughs and admits the rule needs a floor. Nothing was mocked; the rule was just allowed to run.

The tell: Done well, this feels like the ground going soft under your own position, and you cannot point to a step where you were misquoted — that is the honest version, and it means you should look at your rule again. The counterfeit feels different: you say "wait, that's not what I said," and you are right, because a bigger, dumber version of your point got walked off the cliff instead of yours. The other counterfeit ends somewhere merely weird — if the worst thing at the end of the chain is "huh, surprising," nothing has been shown.

The move: The legitimate form: to establish P, temporarily assume its denial; derive from that assumption, together with premises the other party grants, a contradiction or a result they are committed to rejecting; discharge the assumption and conclude P. (The negative form establishes that P is false by assuming P and deriving a contradiction.) The abuse that gets filed as a fallacy substitutes for the opponent's actual claim a stronger or distorted cousin P*, derives absurdity from P*, and reports P as refuted — or derives a result that is merely surprising, unfamiliar, or unwelcome rather than contradictory, and calls that an absurdity.

Why it fails: The form itself does not fail. It is a standard rule of proof with an unbroken record from Euclid onward, and this record classifies it as a technique rather than a defect. What fails are two counterfeits that wear its clothes. The first is the straw-man reductio: the absurdity is derived not from what the opponent asserted but from a strengthened version of it, so the contradiction lands on a position nobody holds while the actual one stands untouched — every step can be impeccable and the refutation still miss its target entirely. The second is the counterfeit absurdity: the derived result is not a contradiction, and not something the opponent has committed to rejecting, but merely something unfamiliar, distasteful, or expensive; a consequence one dislikes is no reason to reject the assumption that produced it, since the assumption may be true and the consequence simply news. A third, narrower failure is logical rather than dialectical: intuitionistic and other non-classical settings license only not-not-P from the untenability of not-P, so the classical form quietly leans on double-negation elimination and cannot be exported to logics that reject it.

• Is the assumption being reduced the other party's actual claim, in their own words, or a strengthened version supplied by the arguer?
• Is the derived result a genuine contradiction, or something the other party has explicitly committed to rejecting — as opposed to merely surprising or unwelcome?
• Does every step from the temporary assumption to the absurd result use premises the other party grants, or is an assumption they would deny smuggled in along the way?
• If the absurdity is real, is it charged to the assumption that actually produced it, or could a different link in the chain be the one that ought to go?
• Is the setting classical? If the argument concludes P from the untenability of not-P, is double-negation elimination available here?

Legitimate when: This is the valid case, and it is the normal case. The technique is sound whenever the derivation from the temporary assumption is correct, the further premises are ones the other party accepts, the terminal result is a genuine contradiction or something they are committed to rejecting, and the blame for the absurdity is correctly assigned to the assumption rather than to some other link. Under those conditions there is nothing to accuse: this is proof by contradiction, among the oldest and most reliable instruments in reasoning.

Ancient: Aristotle treats demonstration through the impossible (hē eis to adynaton apodeixis) as a regular mode of proof and takes as his standing illustration the incommensurability of a square's diagonal with its side: assume the two are commensurable and it follows that the same number is both odd and even, so the assumption must be rejected. Euclid uses the same move at Elements I.6, and the earlier Platonic dialogues raise Socrates' cross-examination of an interlocutor's own admissions into a formal method built on it. The technique is thus among the best-attested in ancient reasoning — and is presented there as proof, never as a defect. — Aristotle, Prior Analytics I.23; Euclid, Elements I.6

Built from: Wikipedia — Reductio ad absurdum · Wikipedia — Proof by contradiction · Logically Fallacious — Reductio ad Absurdum · Stanford Encyclopedia of Philosophy — Intuitionistic Logic · Wikipedia — Straw man

Reductio ad Hitlerum

also known as: reductio ad Hitlerum, playing the Nazi card

Shooting down an idea by pointing out that a monster from history also liked it, without ever saying what is wrong with the idea itself.

The middle-school band parents are arguing in a group chat about whether the kids should wear matching uniform jackets to the spring concert. Renata has been against it for weeks on cost grounds, but tonight she writes: "Matching outfits, everyone in lockstep, no individuality — I mean, we all know which twentieth-century leader loved that look." The chat goes quiet and nobody wants to be the person defending the jackets after that. Nothing has been said about the price of the jackets, whether the school has a budget line, or whether the kids actually mind.

The tell: The room goes cold and you notice you are suddenly more worried about how you look than about whether you are right. There is a swallowing feeling: you know the comparison is ridiculous, and you also know that saying so out loud will sound like you are defending the monster instead of defending the jackets.

The move: A position P is under discussion. The arguer establishes only that Hitler, the Nazi party, or some other maximally reviled figure or regime held P, advocated P, or did something that resembles P, and concludes that P is false, wicked, or not to be adopted. The substitution is of a fact about who once held P for a fact about P itself. A second variant runs at the person: A holds P, Hitler held P, therefore A is morally comparable to Hitler.

Why it fails: Endorsement is not a property of a claim; it is a property of an endorser. A monstrous person's beliefs are not thereby monstrous beliefs, because monsters, like everyone else, hold a mixed bag of true, false, harmless, and terrible views, and nothing about their moral record sorts the bag. The argument therefore offers a fact — Hitler held P — that leaves the probability of P exactly where it was, while borrowing enormous emotional force from the association. It is a named species of guilt by association, and the give-away is that the same form licenses absurd parallels: the regime built highways and promoted early anti-smoking research, so highways and anti-smoking research stand condemned. Worse, the move is self-terminating: once a comparison to the worst crime of the twentieth century is on the table, the actual merits of P can no longer get a hearing, so the fallacy does not merely fail to support its conclusion, it forecloses the inquiry that would settle it.

• Strike every mention of Hitler, the Nazis, or the comparison figure from the passage — does any reason against the position survive?
• Is the connection between the reviled figure and the position merely that both exist, or is a specific mechanism of harm identified?
• Would the arguer accept the same reasoning applied to something the regime plainly got right, such as highway construction or anti-smoking campaigns?
• Is the claim that this position historically produced the atrocity, or only that the same people happened to hold it?

Legitimate when: The comparison is legitimate when the argument turns on a genuine structural feature that carried the harm — the leader principle, the legal stripping of a minority's civil standing, the machinery of a one-party state — and the arguer shows that the proposal at issue shares that feature and would produce that effect. Historical argument from a real causal precedent is evidence; historical argument from shared membership in a reviled person's list of opinions is not.

Built from: Wikipedia — Reductio ad Hitlerum · The Fallacy Files — The Hitler Card · Logically Fallacious — Reductio ad Hitlerum · Internet Encyclopedia of Philosophy — Fallacies

Regression Fallacy

Treating a return to normal as proof that whatever you tried in the meantime is what fixed it.

Twelve-year-old Nico swims the worst 100 free of his life at the county meet — four full seconds off his usual time. His coach, alarmed, spends the next two weeks drilling a new breathing pattern. At the following meet Nico swims a 1:12, right around what he has been swimming all season, and the coach tells the parents on the pool deck that the new pattern clearly did it. Nobody points out that a four-second-off swim was never going to repeat itself, and that the boy did not get faster than his own normal — he just stopped having a terrible day.

The tell: The thing being fixed was at a freak high or freak low right before someone stepped in. If you catch yourself thinking "we had to do something, it was the worst it had ever been," that same sentence is the reason the improvement was coming anyway. The other tell is that the after-number is merely ordinary rather than unusually good.

The move: A quantity X fluctuates around a stable average because each measurement combines a durable component with an unrepeatable one (luck, mood, weather, measurement noise). An unusually extreme value of X is observed. Precisely because it was extreme, an intervention I is applied. The next measurement of X is closer to the average, and the arguer concludes that I produced the change. The extremity of the first measurement is both the trigger for I and, on its own, sufficient to predict the movement attributed to I.

Why it fails: Whenever the correlation between two measurements of the same fluctuating quantity is less than perfect, the value expected on the second measurement lies between the first value and the population average — this is regression toward the mean, a property of imperfectly correlated variables that Galton described in his 1886 study of hereditary stature, not a force acting on anything. An extreme reading is extreme partly because the unrepeatable component happened to run in one direction; that component has no memory, so on the next reading it is far more likely to be ordinary, and the total moves back toward typical whether or not anyone intervenes. The argument's defect is that the no-intervention baseline is silently set at 'the extreme value would have persisted', which is exactly the assumption regression denies. Because interventions are almost always triggered by extremes — you see the doctor when the pain peaks, you install the camera after the crash cluster, you shout at the cadet after the worst landing — this error is systematically generated rather than occasional, and it reliably manufactures evidence for remedies that do nothing and, in the punishment case, evidence against remedies that work. The remedy is structural, not rhetorical: only a comparison group of similarly extreme cases left untreated can separate the effect of I from the movement that was coming anyway.

• Was the intervention applied because the measurement was unusually high or low, rather than at a moment chosen independently of the reading?
• Does this quantity bounce around on its own from week to week, or is it stable enough that an extreme reading would be expected to persist?
• Is there a comparison group of equally extreme cases that got no intervention, and what did those cases do?
• Did the follow-up merely return to the ordinary range, or did it overshoot past the long-run average in the direction the intervention predicts?

Legitimate when: The causal credit is earned when the no-intervention baseline is established rather than assumed: a randomized or matched control of equally extreme cases, a long enough post-intervention series to show the quantity settling below its own historical average rather than merely returning to it, or an interrupted-time-series design that models the expected regression and finds a residual effect on top of it. Regression is a warning about a particular design, not a reason to deny that treatments ever work.

Built from: Wikipedia — Regression toward the mean · Wikipedia — Regression fallacy · The Fallacy Files — Regression Fallacy · The Skeptic's Dictionary — regressive fallacy

Reification

Talking about an idea or a summary as if it were a creature with plans and moods, then arguing from what that creature supposedly wants.

Marisol runs a small candle shop out of her spare room and posts about it online. Over dinner she tells her roommate she has stopped posting entirely: "The Algorithm punished me for going quiet in July, so now it hates my account. There's no point feeding it." Her roommate asks what actually changed — did the reach numbers drop on every post or just the two with no photos? Marisol doesn't know; she has been reasoning about a thing with a grudge instead of about a ranking rule she could test with three posts in a week.

The tell: You notice the argument has a villain or a hero in it who cannot be phoned, met, or asked a question. Somebody is telling you what a word wants. When you try to ask a follow-up — which people, which rule, measured how? — there is nothing there to ask about, and the answer comes back as another sentence about the mood of the thing.

The move: An abstraction A — an average, an institution, a trend, a construct, a collective noun — is a summary of, or a relation among, concrete particulars. The arguer speaks of A as though it were a single concrete thing with a location, a will, appetites, or causal powers of its own: A wants, A demands, A punishes, A has decided. A conclusion is then drawn from those attributed powers, so the argument's support comes from the literalized picture rather than from anything about the particulars A summarizes.

Why it fails: The trouble is a level-of-description confusion. 'The market', 'the average household', 'History', 'the algorithm', 'intelligence' are shorthand for patterns among many individual events and measurements; the shorthand is useful precisely because it drops the detail. Once the shorthand is treated as a concrete agent, it inherits properties nothing in the underlying particulars possesses — an intention, a grudge, a plan — and the argument leans on those invented properties for its force. Whitehead named this the fallacy of misplaced concreteness, the error of mistaking the abstract for the concrete, and its cost is that the claim can no longer be checked: you cannot go and ask History what it demands, whereas you can check what specific historians, voters, buyers, or ranking rules actually did. Stephen Jay Gould's critique of intelligence testing is the standard worked example — treating a score derived from a battery of tests as though it named a single measurable substance in the head, and then reasoning about that substance. The tell is that the argument survives translation into a metaphor but dies on translation back into particulars.

• Is an abstraction being given wants, intentions, demands, rewards, or punishments — verbs that need a mind or a body?
• Rewrite the claim naming only the specific people, rules, or measurements involved: does the argument still deliver its conclusion, or does it now need evidence nobody has offered?
• Could you point to where the named thing is, or check directly what it did, or is it only ever known through the summary?
• Is the personification decorative — a phrase you could delete without changing what is being argued — or is it carrying the weight?

Legitimate when: Personifying shorthand is entirely legitimate as compression or as figure of speech: 'the market rewarded patience', 'the sea was angry', 'the company decided'. It becomes an error only when the literalized picture is doing argumentative work that the underlying particulars cannot do. Likewise, abstractions can be perfectly good subjects of well-defined technical claims — inflation rose, the median fell, the classifier misfired — where each is cashable in a stated procedure over concrete data.

Built from: Wikipedia — Reification (fallacy) · Stanford Encyclopedia of Philosophy — Alfred North Whitehead · Internet Encyclopedia of Philosophy — Fallacies (Hypostatization) · Logically Fallacious — Reification

Relative Privation

also known as: appeal to worse problems, not as bad as

Waving away someone's problem by pointing at a bigger problem somewhere else, as if the bigger one made the smaller one stop existing.

Halfway through a Friday dinner rush, Theo tells the kitchen manager that the walk-in cooler has been running warm for three days and they're throwing out produce. The manager doesn't look up from the ticket rail: "Theo. There are restaurants in this town that closed last year. People would kill for these hours. Warm lettuce is not a problem." Theo goes back to the line, the cooler stays warm, and nothing about the compressor has been discussed. Two other restaurants closing did not change the temperature of anything.

The tell: You feel ashamed for having spoken, and you notice the shame arrived before any answer did. There is a small stuck feeling: the thing you raised is still exactly as broken as it was thirty seconds ago, but now you would look ungrateful for mentioning it again.

The move: A raises a problem or complaint P. B observes that some other problem Q is worse than P — often much worse, and often somewhere else entirely — and concludes that P is not a real problem, does not merit attention or resources, or that A has no standing to raise it. The comparison to Q is offered as the whole reason, with no claim that addressing P and addressing Q are in competition for the same resource. The mirror form runs upward: P is compared to an ideal best case and dismissed as inadequate.

Why it fails: The severity of a problem is not a comparative property. That someone elsewhere walks four miles for drinking water does not make the sediment in this building's tap water safe, does not make the landlord's obligation smaller, and does not change any fact about the pipes. The comparison changes only how the complaint feels. The move borrows its plausibility from a real principle — that scarce resources should go to the greatest need first — and applies it in cases where nothing is scarce and no allocation is being decided; attention, sympathy, and repair are not usually zero-sum between a Nairobi water crisis and a broken filter in Toledo. The most efficient way to see the defect is that it generalizes: if a problem is dismissible whenever a worse one exists, then every problem in the world is dismissible except exactly one, which is a standard nobody actually holds. And the dismissal is typically aimed at the complainant's standing rather than at the complaint, which is why it wounds rather than informs.

• Is the question on the table which problem to fix first, or whether this problem counts at all?
• Are the two problems genuinely competing for the same budget, hours, or attention, or could both be addressed?
• Does the comparison change any fact about the original problem — its cause, its cost, its remedy — or only how bad it is allowed to feel?
• Applied consistently, would this reasoning dismiss every complaint on earth except the single worst one?

Legitimate when: Comparative severity is exactly the right consideration in genuine triage, where the resource really is rivalrous and prioritization is the actual question: an emergency department with one surgeon, a maintenance budget that funds three of eleven repairs, a legislature with room for two bills. The legitimate form ends in 'so this one waits', with a stated reason why the two compete; the failed form ends in 'so this one doesn't count'.

Built from: Wikipedia — Fallacy of relative privation · Logically Fallacious — Relative Privation · Internet Encyclopedia of Philosophy — Fallacies

Retrogressive Causation

Reaching for more of the exact thing that caused the trouble, and calling that the way out of the trouble.

Marcus has been exhausted for two weeks and he knows why: he scrolls in bed until two in the morning. His sister asks what he's going to do about it and he says, completely sincerely, "I'll just stay on my phone until I actually get sleepy — then I'll fall asleep instantly." He does, and at 2:40 a.m. he is wider awake than when he started. The phone is the reason he isn't sleepy, and the plan for getting sleepy is more phone.

The tell: The plan sounds oddly comfortable — it asks nothing of you that you weren't already doing. Then a small back-of-the-neck itch: you have run this experiment before, many times, and every previous run left you deeper in the thing you were trying to get out of. If a plan is indistinguishable from the problem, that is the itch.

The move: X is the cause, or a major cause, of unwanted effect Y. Facing Y, the arguer proposes more of X as the remedy — reasoning that because X and Y are already bound together, applying X will dissolve Y. The recommendation reverses the direction of the causal arrow it has just acknowledged: the thing identified as the source of the problem is enlisted as the solution to the problem it sources, so the proposal, if acted on, feeds the very process it claims to interrupt.

Why it fails: A cause does not become a remedy by being close at hand. If X produces Y, then applying more X has an expected effect on Y that is positive, not negative, and the argument offers nothing that would flip that sign — no threshold, no saturation point, no counteracting mechanism, only the association between X and Y as though association alone were steerable in either direction. What generates the mistake is usually that X delivers immediate relief from the *feeling* of Y while enlarging Y itself: another drink quiets the shame the drinking caused, another late-night scroll pushes off the tiredness that late-night scrolling produced, another status meeting relieves the anxiety that too many status meetings created. The short-run comfort is real, which is what makes the reasoning feel supported, but it is evidence about the symptom's timing rather than about the effect's size. The result is a loop that reads as a strategy: each application of the remedy generates the condition that calls for the next one, and the proposal survives precisely because its failure looks like a reason to apply it again.

• Has the arguer already granted, in this same passage, that the proposed remedy is what produces the problem?
• If the remedy were applied ten more times, would the underlying problem plausibly be smaller, or larger?
• What mechanism is offered for why more of the cause would reduce rather than increase the effect, beyond the fact that it brings immediate relief?
• Is the relief being promised relief from the problem itself, or only relief from noticing it?

Legitimate when: Some genuine remedies really do run through the agent that caused the harm, and those are not this error: a controlled dose of the allergen in immunotherapy, a tapering schedule that supplies the substance precisely to prevent withdrawal, a hair-of-the-dog protocol under medical supervision, or a stimulus that provokes a compensating response with a stated dose-response curve. What separates them is that a mechanism and a quantity are specified, the dose is bounded and declining, and the claim is testable against a no-treatment case — rather than 'more of the same, indefinitely, because the same is what we have'.

Built from: Logically Fallacious — Retrogressive Causation · Internet Encyclopedia of Philosophy — Fallacies · Stanford Encyclopedia of Philosophy — Backward Causation

Righteousness Fallacy

Deciding somebody must be right about the facts because they are a good person with their heart in the right place.

At the dog rescue's Saturday meeting, a volunteer named Pearl says the shelter should switch every dog to a raw-meat diet, including the two with kidney trouble. When someone asks whether the vet has weighed in, three people jump in at once: Pearl has fostered eighty dogs, she drives to the border twice a year at her own expense, she once slept in her car with a litter of puppies. All of that is true, and none of it is a reason to believe that raw meat is safe for a dog in kidney failure. The vet still hasn't been asked.

The tell: You want to check the claim and you can feel that checking it would be read as an attack on a person everyone loves. The praise arrives instead of an answer, and it arrives fast. If asking "how do we know that?" would make you look ungrateful rather than careful, this is the move you are up against.

The move: S asserts a claim P about some matter of fact. The only support offered for P is that S is a good person — kind, generous, sincere, well-intentioned, morally serious. From S's virtue the arguer concludes that P is true, or that P need not be checked. Moral standing is substituted for subject-matter competence, and a judgment about a person is converted into a judgment about a claim.

Why it fails: Goodness and accuracy are separately acquired. A person becomes generous through disposition and habit and becomes right about soil chemistry, traffic modelling, or a dog's diet through study, practice, and correction — and neither process supplies the other. Virtue is therefore not evidence for P; it merely makes the assertion of P feel trustworthy, which is a fact about the hearer's response rather than about the world. The move is the mirror image of the abusive ad hominem: where that one infers falsity from the arguer's badness, this one infers truth from their goodness, and both are the same substitution of the speaker for the claim, differing only in sign. There is a narrow legitimate neighbour that lends it cover — honesty does bear on whether a witness is reporting sincerely, so character is relevant to testimony about what the speaker themselves saw or did. But that relevance is confined to sincerity, and it collapses the moment the claim requires expertise instead of eyewitness honesty, because a scrupulously sincere person can be scrupulously wrong.

• Strip out every mention of the speaker's character — is any evidence for the claim left standing?
• Is the claim a report of what the speaker personally witnessed, or a technical judgment requiring training the speaker's virtue does not supply?
• Would the same claim, with the same supporting detail, be accepted from someone the group dislikes?
• Is the appeal to sincerity — she is not lying — or to accuracy, and is the argument sliding from the first to the second?

Legitimate when: Character legitimately bears on credibility in a bounded way: when the question is whether a first-hand report is honest, a track record of truthfulness raises the probability that the reporter is not fabricating, which courts and ordinary life both rely on. It is also fine to draw moral conclusions from moral character — she is trustworthy with the cash box. The line is crossed when honesty is treated as competence, so that being good about telling the truth becomes being right about the subject.

Built from: Logically Fallacious — Righteousness Fallacy · Logically Fallacious — Self-Righteousness Fallacy · Internet Encyclopedia of Philosophy — Fallacies · Wikipedia — Argument from authority

Rights To Ought Fallacy

Treating "I'm allowed to do this" as if it settled the question of whether doing it was a good idea.

The design pod shares one long desk, and Kwame has started taking calls on speaker with his headphones sitting right there next to the keyboard. When Rosa finally asks if he'd mind using the headset, he shrugs: "It's an open office. There's no rule against speaker. I'm within my rights here." He is — there is genuinely no policy about it. What Rosa asked was whether four people can hear his dentist confirming an appointment while they're trying to write, and that question has not been touched.

The tell: You asked whether something was a good idea and got an answer about what's permitted, delivered like a door closing. Notice how quickly the conversation reframes you as the one trying to take something away — you wanted to talk about the noise, and now you are apparently against freedom.

The move: It is established that A has a right — legal, constitutional, or moral — to do phi; in Hohfeld's terms, A has a privilege, meaning A is under no duty not to phi. From this the arguer concludes that A ought to phi, that phi-ing is good or admirable, or that phi-ing is immune from criticism. The word doing the work shifts silently across the argument, from 'may' in what was shown to 'should' in what is claimed.

Why it fails: Permissibility and advisability are two different normative categories, and rights mark only the first. A privilege-right is defined negatively — the absence of a duty against the act, and a protection against being compelled otherwise — which tells you what may not be done *to* A, not what A has reason to do. The gap is not exotic; it is the ordinary space in which most of etiquette, prudence, and decency lives, and philosophers call the inhabitants of that space rights to do wrong: you have every right to tell a grieving friend exactly what you think of her late husband, and you should not. Because the two senses share the vocabulary of 'right' — one meaning entitlement, the other meaning correct — the slide can pass for a restatement rather than an addition. The argument therefore needs an unstated bridge, that whatever is permitted is what should be done, which is false on a single counterexample and which no one defends once it is written out. And it does real damage in argument, since it converts every criticism of a choice into an alleged attack on the freedom to make it, and so ends the discussion of whether the choice was any good.

• Is the thing established that the act is allowed, while the thing claimed is that the act is good or advisable?
• Write out the missing bridge — 'whatever one is permitted to do, one should do' — and ask whether the arguer would sign it.
• Can you name an act the speaker agrees people have every right to do and nonetheless should not do?
• Is criticism of the act being answered as though it were a demand to prohibit the act?

Legitimate when: The step from a right to a recommendation is fine when a separate reason supplies it: 'you have the right to appeal and you should, because the deadline is Friday and your case is strong' argues from the merits, with the right merely establishing that the option is open. Rights also legitimately settle questions about what third parties may do — you may not be fired for it, you may not be silenced — which is a different conclusion from 'and therefore it was the right call'.

Built from: Logically Fallacious — Rights To Ought Fallacy · Stanford Encyclopedia of Philosophy — Rights · Internet Encyclopedia of Philosophy — Fallacies

Scapegoating

Pinning a whole problem on whoever is easiest to blame, and calling that an explanation of what went wrong.

Tickets at the diner have been running twenty minutes long for two months. At the Sunday staff meeting the owner announces he has figured it out: it's Ruben, the dishwasher who started in March. "Ever since that kid got here, nothing comes out on time." Ruben is nineteen, works alone, speaks less English than anyone else in the room, and is the only person there without an ally. Nobody mentions that the second fryer died in February and the line has been running on one, which is what everyone at the pass has been saying since February.

The tell: The explanation arrives already finished, and it lands on the person in the room with the least ability to answer back. Watch for relief in the tone — people sound lighter after it is said. And notice what stops being discussed the moment the name is spoken: usually the expensive thing, the thing the person doing the blaming would have to fix.

The move: A problem P persists and its causes are diffuse, structural, or uncomfortable to name. A person or group G is available — new, unpopular, low-status, outnumbered, or unable to answer back. The arguer assigns responsibility for P to G and concludes that removing, punishing, or restricting G will resolve P. The support offered for the causal claim is G's availability and unpopularity, not any evidence tying G's conduct to P.

Why it fails: Two defects run together, which is what makes the move so durable. The first is a relevance failure: that a group is disliked or convenient bears on how satisfying it is to blame them and not at all on whether they caused anything, so the sentence 'nobody likes G, therefore G is responsible' has no content. The second is causal: even where G contributes something, the arguer collapses a many-factor problem onto one actor, which both overstates G's contribution and hides every other cause — including, characteristically, the arguer's own. The test that exposes it is counterfactual and cheap: remove G entirely and ask whether P persists. Where P is driven by a supply constraint, a broken process, or a decision made two levels up, the answer is that it persists exactly as before, and the removal buys nothing but the feeling that something was done. That feeling is the mechanism: the move meets a real appetite for the sense that accounts have been settled, which is why it survives its own repeated failure and why the next candidate is nominated as soon as the first is gone.

• What evidence connects this person or group to the problem, other than that they are disliked, new, outnumbered, or convenient?
• If they were removed tomorrow, would the problem actually stop — and has anyone checked what happened last time someone was blamed for it?
• Which structural or multi-cause explanations does the blame push out of the conversation, and who benefits from their absence?
• Is the blamed party in a position to answer the charge, and has anyone asked them?

Legitimate when: Assigning responsibility to a specific person or group is entirely proper when the evidence identifies them: an audit trail, a decision they made, a documented pattern of conduct, a mechanism connecting what they did to the harm that followed. Accountability is not scapegoating. The distinguishing marks are that the evidence came before the accusation rather than after the choice of target, that the accused can answer, and that the account leaves room for the other causes the record shows.

Built from: Wikipedia — Scapegoating · Logically Fallacious — Scapegoating · Internet Encyclopedia of Philosophy — Fallacies

Selective Attention

Building a case out of the times you were right and simply never counting the times you weren't.

Bridget has told her running club for a year that a particular shoe brand causes injuries. She can name five people who got hurt in them — she remembers each name, each race, each injury. What she does not have on the list is the eleven other members who have run in the same shoe for two seasons with nothing worse than a lost toenail, or the four people who blew out a hamstring in a different brand entirely. She never went looking for those; the ones who got hurt in that shoe are simply the stories she noticed and repeated, and after a year the list looks like data.

The tell: Every single example you're given checks out, and the overall picture still feels tilted. Ask yourself what a case *against* their claim would look like and whether they could name even one — if they can't, and they've never gone looking, you are being shown a filtered set of true things rather than the whole set.

The move: A body of evidence E bears on a claim C and divides into a favourable part E+ and an unfavourable part E−. The arguer notices, remembers, and cites the members of E+ while leaving E− unsought, unmentioned, or discounted, and then concludes C on the strength of the surviving sample. Every cited instance may be genuine; what is defective is the filter that decided which instances reached the argument.

Why it fails: An induction is only as good as the procedure that produced its sample, and this procedure is not blind to the answer — it selects on the very thing being tested. The cited instances therefore cannot discriminate between 'C is true' and 'C is false but confirming cases exist', because a false C with a decent number of coincidences produces exactly the same visible record as a true one. The failure is a violation of what informal logicians call the total evidence requirement: an argument must rest on all the relevant evidence available to the arguer, since a conclusion licensed by a subset can be reversed by the remainder. Two features make it especially hard to catch from the inside. First, the filtering is often unconscious — the attention itself is biased, so nothing feels omitted — and second, the arguer can honestly say every claim they made was true, which is exactly compatible with the overall picture being backwards. Cicero preserves the ancient version of the point in Diagoras and the votive tablets, where the shipwreck survivors who prayed are all hanging on the temple wall and the ones who drowned are not painted anywhere.

• What would a disconfirming case look like, and was any effort made to look for one?
• Was the evidence gathered by a route that could have turned up the unfavourable cases — or does the route only ever pass the favourable ones?
• Does the favourable and unfavourable record together, counted rather than recalled, still support the claim?
• Of the specific points the other side raised, which were answered and which were passed over in silence?

Legitimate when: Setting evidence aside is not only permitted but required when the evidence is irrelevant, unreliable, or already superseded — a good argument is selective in that sense, and answering every point ever raised is not a standard anyone meets. What makes the selection legitimate is that the criterion is stated and independent of which side the evidence favours: excluded because the measurement was uncalibrated, because the sample was contaminated, because the point does not bear on the claim. When the only thing the excluded items have in common is that they told against the conclusion, the criterion was the conclusion.

Ancient: Cicero reports the story of Diagoras of Melos, who was shown the votive tablets in a temple — offerings from sailors who prayed in a storm and survived — and invited to admit the gods' power. Diagoras answered that there are nowhere any pictures of those who prayed and were drowned. The anecdote isolates the exact defect: the evidence on the wall was collected by a procedure only survivors could pass through, so it could not distinguish answered prayer from ordinary luck. — Cicero, De Natura Deorum III.89

Built from: The Fallacy Files — One-Sidedness · Wikipedia — Cherry picking · Logically Fallacious — Selective Attention · Wikipedia — Diagoras of Melos

Self-Righteousness Fallacy

Defending what you said or did by pointing at how good your motives were, as if meaning well settled whether you were right.

Dani reads her younger brother Owen's journal while he's at practice, then confronts him about a paragraph in it. When he asks what gave her the right, she says, "I did it because I love you and I was scared for you. My heart was in the right place, so don't act like I did something wrong." Her worry is completely genuine. It is also entirely compatible with reading a seventeen-year-old's journal having been the wrong way to act on it, and that second question never gets discussed.

The tell: You asked about the thing that happened and got an answer about the other person's feelings. There is a strange narrowing in the room: any further question now sounds like you are accusing them of being a bad person, when all you wanted to know was whether the thing worked or whether it was okay to do.

The move: The arguer asserts a claim P, or defends an action A that they took, and offers as support a fact about their own inner life: my motives were pure, I meant well, my conscience is clear, I would never do such a thing. From the purity of the intention the arguer concludes that P is true, or that A was the right thing to do. The first-person report of good intention is put in the place where evidence about the claim, or about the action's actual effects, belongs.

Why it fails: Intention and outcome are logically independent. Meaning well is a fact about the state of the agent before and during the act; whether the act helped, whether the claim is accurate, are facts about the world afterward, and nothing carries you from the first to the second — the entire vocabulary of the well-meant disaster exists because people constantly do harm with clean hearts. As self-defence the move is additionally circular in a way the other-directed version is not: the only witness to the purity of the motive is the person whose conduct is in question, so the offered evidence is inaccessible to everyone else and cannot be weighed, which converts the dispute from one about the record into one about the speaker's inner life, where the speaker cannot lose. It also equivocates across two different verdicts that hearers naturally conflate — moral credit for trying and correctness of the result — so that granting the first, which is often deserved, is taken to have granted the second, which is what was actually being contested.

• Is the support offered a fact about the speaker's intentions rather than about the claim or the outcome?
• Could the very same good intentions have coexisted with the bad result being alleged — is the defence compatible with the charge?
• Would the speaker accept this defence from an opponent who said the identical sentence about themselves?
• Is the argument distinguishing credit for trying from the question of whether the thing worked, or quietly merging them?

Legitimate when: Intention is genuinely decisive in some questions, and citing it there is not an error: whether an act was malicious, negligent, or accidental turns on the agent's state of mind, and moral and legal judgments of culpability rightly hinge on it. It is also honest to say 'I acted in good faith' as a plea about blame. The move fails only when good faith is offered as evidence that the claim is accurate or the outcome good — that is, when a question about the world is answered with a report about the self.

Built from: Logically Fallacious — Self-Righteousness Fallacy · Logically Fallacious — Righteousness Fallacy · Internet Encyclopedia of Philosophy — Fallacies · Wikipedia — Argument from authority

Self-Sealing Argument

Setting a claim up so that every possible outcome counts as proof of it, and nothing on earth could ever count against it.

Tamsin has the group chat on day four of a juice cleanse and the reports are coming in. Jo feels great — "see, that's the cleanse working." Priya has a splitting headache and feels awful — "that's the toxins leaving, it's working even harder on you." Ellis says he feels completely normal, and Tamsin says that means it's working so gently he can't tell. There is no message any of them could send back that would show up as bad news for the cleanse.

The tell: You realize, a beat late, that you could have said the opposite thing and gotten the same answer. Every objection you raise gets converted into more support for their side, and disagreeing with them starts being treated as a symptom of the very thing you're disagreeing about. You feel less like you're talking and more like you're being read.

The move: A position P is advanced together with an interpretive rule R that assigns every possible observation to the credit of P: confirming evidence supports P directly, and disconfirming evidence E is re-described by R as a further symptom, a cover-up, a sign of resistance, or proof of how deep the problem runs. The arguer concludes that nothing counts against P, and treats that immunity as strength. Denial by an opponent is itself absorbed as confirmation, so the position's supporters can never lose an exchange.

Why it fails: A claim earns support by ruling something out. If P is compatible with every outcome — the polls up, the polls down, the patient improving, the patient worsening — then observing the world tells you nothing about P, and the accumulated 'confirmations' are not evidence but restatements of the interpretive rule. The defect is therefore not that P is false; a self-sealed position may happen to be true. It is that the argument offered for P has no evidential content, because the rule R was chosen precisely so that it could not fail, and support that cannot fail is not support. Popper's cases are the standard illustration: he objected to psychoanalysis not because its claims were absurd but because it did not exclude anything, and to the Marxist defenders of his day because they answered refutation with reinterpretation of both theory and evidence, a 'conventionalist twist' that saved the theory at the cost of its standing. The move is corrosive in argument because it converts the opponent's every move into fuel — objecting shows you are repressing it, staying silent shows you have no answer — which is why the honest question to put to it is not 'what supports this?' but 'what would count against it?'

• Ask the arguer to name one observation they would accept as counting against the claim — do they produce a real one, or none at all?
• Are objections being answered with evidence, or re-described as symptoms, resistance, corruption, or proof of how deep the problem goes?
• Would the opposite of the current situation also have been read as confirming the claim?
• Is your own denial being counted as support for the charge against you?

Legitimate when: Some claims are legitimately hard to test and it is not an error to hold them: definitions, mathematical truths, and personal commitments are not in the business of being refuted by observation, and it is perfectly honest to say 'this is what I believe, and I am not offering it as evidence'. Well-supported theories also properly resist single anomalies, since a lone contrary result is more often a bad measurement than a refutation. The line is drawn by whether the arguer can state, in advance, what pattern of results would make them give the claim up — and whether they have ever accepted such a result.

Built from: Logically Fallacious — Self-Sealing Argument · Wikipedia — Falsifiability · Internet Encyclopedia of Philosophy — Fallacies

Shifting of the Burden of Proof

Making a claim you never backed up, then telling the other person it stands until they can prove it wrong.

Eighty dollars in rent cash goes missing from the kitchen jar, and Colby tells the house he's fairly sure Nadia took it. Nadia asks what makes him think that. "Well, can you prove you didn't?" he says. She can't — nobody can prove they didn't take cash out of a jar in an empty kitchen — and by Thursday the story around the house is that she probably did, on the strength of a question she was never in a position to answer.

The tell: You notice you're the one doing all the work, and you never made a claim. The other person hasn't given you a single reason for what they said, and yet somehow you're the one scrambling to produce receipts. There's also a sinking recognition that what they're asking for doesn't exist — you cannot produce evidence of a thing not happening.

The move: A asserts P, a claim that departs from the standing presumption and for which no support has been given. Challenged, A does not supply support but demands that the challenger establish not-P. The challenger cannot, or does not, and A treats that failure as grounds for P — or simply as licence to go on asserting P until disproved. The obligation to argue has been transferred from the party who introduced the claim to the party who merely declined to accept it.

Why it fails: Argument runs on an allocation of obligations, and the allocation is not arbitrary: the one who asserts must prove, not the one who denies, as Roman law had it and as every deliberative practice since has needed. The reason is practical and unavoidable — the space of unsupported claims is infinite, refuting each one costs real effort, and a rule permitting the burden to be reversed at will would let anyone install any claim as the default at zero cost, which is a procedure for generating belief rather than for testing it. Once reversed, the argument also runs on the wrong currency: failure to refute is not support, it is silence, and treating silence as evidence is the argument from ignorance with the direction of the demand made explicit. The reversal is additionally asymmetric in a way that is easy to miss in the moment — a general negative is often unprovable in principle, so the demand is not merely unfair but unmeetable, which is precisely why it is attractive to the party making it. What the challenger owes is not a disproof but a reason for the presumption they are standing on.

• Who put the claim into the conversation, and what support did they offer before anyone asked them to?
• Is the demand 'show me why you believe this' or 'show me why I shouldn't' — and is failure to meet the second being counted as a point scored?
• Is the disproof being demanded even obtainable, or is it a general negative nobody could establish?
• What does the governing setting — a court, a lab, a clinical review, a friendly argument — actually say about who carries the burden here, and does the claim depart from that default?

Legitimate when: Burdens genuinely do shift, and saying so is not this error. Once a party has made a prima facie case, the burden of going forward passes to the other side, which is ordinary and correct; a settled default — the presumption of innocence, an established scientific consensus, a signed contract's plain terms — legitimately places the burden on whoever wants to overturn it; and a thorough search that came up empty is real negative evidence rather than mere absence. The move fails only when a party who introduced an unsupported departure from the default demands refutation instead of providing support.

Ancient: The jurist Paulus states the rule as: Ei incumbit probatio qui dicit, non qui negat — proof lies upon the one who asserts, not upon the one who denies. Preserved in the Digest compiled under Justinian in the sixth century, it is the classical statement of the burden-allocation norm this move violates, and the ancestor of the presumption of innocence in later law. — Justinian, Digest 22.3.2 (Paulus, Ad Edictum, book 69)

Built from: Wikipedia — Burden of proof (philosophy) · Wikipedia — Argument from ignorance · Logically Fallacious — Shifting of the Burden of Proof · The Latin Library — Digest of Justinian, Book XXII · Internet Encyclopedia of Philosophy — Fallacies

Shoehorning

Cramming whatever just happened into your favourite explanation, no matter what shape it was to begin with.

At Sunday dinner, Aunt Ines has an account of everyone's week and it is always the same account. Devon's car wouldn't start: Mercury retrograde. Her sister's flight was delayed: Mercury retrograde, obviously. When Devon points out that his brother had a great week and got the apartment he wanted, she says that's the retrograde too — old things coming back around. Nobody at the table can think of a week that would have counted as the ordinary kind.

The tell: You catch yourself thinking "they'd have said that either way." Every new thing that happens somehow lands in the same story, and the details that don't quite fit get quietly sanded down rather than explained. It stops feeling like they're figuring out what happened and starts feeling like they're filing it.

The move: The arguer holds a favoured framework F — a personal theory, a political programme, a religious reading, a pet diagnosis — settled in advance of the case at hand. A new event or piece of evidence E arrives. The arguer force-fits E into F, reinterpreting whichever features of E resist the fit, and presents the resulting fit as confirmation of F. The applicability of F to E is presupposed rather than argued, and no version of E is admitted under which F would have been the wrong lens.

Why it fails: The presumption smuggled in is that F is the right frame for E, and that is exactly what an argument would have to establish. Once F is assumed applicable, fitting E into it is a matter of interpretive effort rather than evidence — a sufficiently flexible framework can be made to accommodate almost anything, so the fit demonstrates the framework's elasticity rather than its truth. The second defect is that no rival is ever put alongside it: the case for F consists in F's being able to explain E, with no comparison to the explanation a competing frame would give and no accounting for the features of E that had to be bent. That is why a text written two thousand years ago can be read as predicting last month's oil spill, and why the same quarterly numbers can be read as vindicating a strategy whether they went up or down. The give-away is prospective: ask what E would have had to look like for the arguer to say 'this one doesn't fit my framework', and there is usually no answer, which means the framework was never at risk and the case was decided before the evidence arrived.

• Was the framework in place before this event, and did the arguer commit in advance to what would count as a poor fit?
• Which specific features of the event had to be stretched, downplayed, or reinterpreted to make the fit work?
• Has any rival explanation been put beside this one and compared, or is the framework the only candidate on the table?
• Would the opposite outcome have been folded into the same framework just as readily?

Legitimate when: Applying a general framework to a new case is the ordinary business of diagnosis, forecasting, law, and science, and it is legitimate when the framework earns its application: when the case exhibits the features the framework specifies, when the fit was predicted before the outcome was known rather than constructed afterward, when the awkward details are stated rather than smoothed away, and when a rival account has been considered and found to fit worse. A framework that can be shown in advance to have been wrong about some cases and right about this one is doing work; one that fits everything is doing none.

Built from: Logically Fallacious — Shoehorning · The Fallacy Files — One-Sidedness · Wikipedia — Falsifiability · Internet Encyclopedia of Philosophy — Fallacies

Slippery Slope

also known as: thin end of the wedge, camel's nose

It claims one small step will snowball all the way to disaster, without ever showing why it has to.

A dad tells his teenager, 'If I let you stay out until 11 tonight, next week you'll want midnight, then you'll want to stay out all night, then you'll stop listening to any curfew at all.' The kid just asked for one extra hour tonight — nothing has actually shown that tonight's hour leads to any of the rest.

The tell: The conversation suddenly jumps from a small, specific ask straight to the worst possible ending, skipping over every place things could realistically stop along the way — and nobody ever explains why step one actually forces step two to happen.

The move: A₁ (a modest, often reasonable first step) will lead to A₂, which will lead to A₃, and so on down a chain to Aₙ, an extreme and clearly unacceptable outcome; because Aₙ must be avoided, A₁ itself must be rejected — with the whole chain asserted rather than established, transition by transition.

Why it fails: Each link in the chain (A₁→A₂, A₂→A₃, …) is treated as automatic or near-certain when no causal mechanism, comparative precedent, or absence-of-stopping-point has actually been shown for that specific transition. Even if every individual step sounds superficially plausible, the probability of reaching the terminal outcome is the product of all the transition probabilities, which can be vanishingly small even when each step feels intuitive in isolation. The argument borrows its persuasive force from vividness (a scary ending) rather than from any demonstrated chain of causation, and it silently ignores legal, institutional, physical, or psychological barriers that could halt the slide at any point.

• For each individual transition (A₁→A₂, A₂→A₃, …), what specific evidence — a documented mechanism, legal precedent, or comparable past case — supports it, versus it just sounding plausible?
• Is there a stopping point between here and the terminal outcome (a law, a cost, a backlash, an institutional check, a physical limit) that the argument never addresses?
• If you multiply the probability of each transition actually happening, how likely is the terminal outcome really — and does that match how inevitable the argument makes it feel?
• Can you find real cases where step one happened and the chain simply stopped partway, never reaching the feared endpoint?

Legitimate when: The chain is a genuinely evidenced causal, legal, or institutional sequence — each transition independently supported by mechanism, precedent, or data (e.g., documented cases of a specific legal exception being cited to justify broader exceptions, or a well-studied behavioral priming effect) — and the cumulative probability of the terminal outcome, once the transitions are actually multiplied out, remains high enough to be decision-relevant. In that case the argument functions legitimately to shift a burden of proof rather than to smuggle in an unsupported chain.

Ancient: In his dialogue on friendship, Cicero has the speaker Laelius reflect on Rome's political decline in terms of a slope: matters glide readily down the path of ruin once they have taken a start, an image later writers on the fallacy point to as the earliest recorded use of the slope metaphor for this kind of chained-decline reasoning. — Cicero, Laelius de Amicitia (XII.41, 44 BC)

Built from: Wikipedia — Slippery slope · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies

Special Pleading

Insisting on a rule for everyone else and then quietly writing yourself an exception you never have to justify.

The no-phones-at-dinner rule was Ana's idea, and she enforces it hard — the kids' phones go in a basket by the door, no negotiating. On Wednesday her own phone buzzes in her lap and she answers it at the table, mid-conversation. Her fourteen-year-old raises an eyebrow and she says, without looking up, "This is work, mija. It's different." It's not, particularly: half the point of the basket was that dinner is the one hour nobody's attention goes elsewhere, and work is exactly the thing pulling hers away.

The tell: Someone says "this is different" and then just... stops, as if the sentence had finished itself. Notice that no reason follows — the difference is announced, not explained. And notice you can predict, instantly, that the same excuse from you would not have been accepted.

The move: A standard S is accepted and applied to a class of cases. The arguer's own case, or a favoured case, falls squarely within that class. Rather than accept S or argue against S generally, the arguer claims an exemption for the favoured case and grounds it in a feature that has no bearing on why S exists — I am busy, I understand the situation, my reasons are good, this one is different. The exemption is asserted, and the burden of showing a relevant difference is skipped.

Why it fails: A standard is not a bare list of prohibited cases; it exists for a reason, and an exception is legitimate exactly when the reason does not reach the case. That is why guide dogs are admitted to restaurants that bar animals: the rule is there for hygiene and disruption, a trained service animal does not raise those concerns, and the blind patron would otherwise be shut out — so the reason for the rule itself carves the exception. Special pleading skips this test and offers a feature the reason cannot use, so the exemption rests on the arguer's preference for the favoured case rather than on anything about the case. Two things follow. The argument is inconsistent, since the arguer both endorses S and does not apply it, and there is no third proposition reconciling the two. And it is unstable under exchange: the exempting feature — I have a lot going on, my intentions are good, I know what I'm doing — is available to every person the standard has ever bound, so accepting it would dissolve the standard entirely, which is a result the arguer explicitly does not want. The reversal test is therefore diagnostic, not merely rhetorical: ask whether the exemption would be granted to a rival citing the identical ground.

• What relevant difference is offered between the favoured case and the cases the standard binds, beyond the fact that this one is the arguer's?
• Does the stated difference connect to the reason the standard exists, or only to the arguer's convenience or good opinion of themselves?
• Would the arguer grant the same exemption, on the same ground, to someone they disagree with?
• If everyone who could truthfully cite this ground were exempted, would the standard have anything left to govern?

Legitimate when: Principled exceptions are ordinary and good: rules routinely admit them where the rule's own rationale runs out, and stating one is an argument rather than an evasion. What makes it principled is that the difference is spelled out, it connects to the purpose of the standard, and it is stated in general terms so that anyone else in the same position gets the same exemption. It is also perfectly legitimate to argue that the standard itself is wrong — that is a different and honest move, provided the arguer then drops it for everyone rather than only for themselves.

Built from: Wikipedia — Special pleading · The Fallacy Files — Special Pleading · Internet Encyclopedia of Philosophy — Fallacies · Logically Fallacious — Special Pleading

Spin Doctoring

Describing something so carefully and so favourably that people walk away with the opposite impression of what actually happened, without a single false word being said.

Marguerite is showing an apartment above a bar on a busy street. The listing calls it "a lively, characterful walk-up in the heart of the district — vintage fixtures throughout, steps from the city's best nightlife." On the tour she mentions that the radiator is original to the building and that the neighbours are "very social people." Everything she says is true. The apartment is a fourth-floor unit with no elevator, a heating system from 1958, and a bass line coming through the floor until two in the morning, and none of that is anywhere in the sentence.

The tell: You leave feeling reassured and you can't say what reassured you. Try repeating back what you were told in your own flat words — if the summary comes out much worse than the original did, the difference was the wording. The other tell is grammar: bad things keep happening with nobody doing them.

The move: A body of facts F bears unfavourably on the speaker's interest. Rather than dispute F, the speaker re-presents it: selecting the favourable items, describing the unfavourable ones in euphemism or in the passive voice, moving the comparison baseline, and attaching evaluative wording that carries a verdict. The audience is invited to draw conclusion C from the re-presentation, and every individual statement made can survive a fact-check. The support for C comes from the framing rather than from anything the facts establish.

Why it fails: The wording is not the evidence, and here the wording is all there is. Because the reframing is engineered to be literally defensible, it cannot be answered by contradiction — the shortfall really was found, the review really did identify accounting practices — and yet the picture the audience walks away with is the reverse of the one the facts support. That is the specific defect: evaluative and selective framing is offered where a reason bearing on the claim belongs, so the audience's judgment is moved without their evidence being changed. Two mechanisms do most of the work. Selection sets what the audience knows about, since a favourable subset presented as the whole is indistinguishable, from the outside, from a favourable whole. And euphemism plus agentless grammar detach events from responsibility — 'mistakes were made' names an occurrence with nobody in it. The clean test is substitution: restate the identical facts in flat, neutral language, name the actors, and restore the omitted items, then ask whether the conclusion still follows. Where it evaporates, the framing was carrying the argument.

• Strip out every evaluative adjective and restate the facts flatly — does the favourable conclusion survive?
• Who is the actor in each sentence, and have the unfavourable sentences been written so that nobody performed them?
• What is the comparison baseline, and was it chosen because it flatters — a different year, a projection, an unusually bad quarter?
• Could the same underlying facts be arranged to support the opposite reading just as easily, using only true statements?

Legitimate when: Choosing how to present true information is not itself dishonest — every explanation selects, orders, and characterizes, and advocacy in a setting where both sides speak is a legitimate and expected practice. Framing is defensible when the selection criterion is stated or obvious, when the unfavourable material is present rather than absent, when the actors are named, and when the audience has access to the underlying record. It becomes this error when the presentation is designed so that a reasonable audience would form a belief the full record does not support.

Ancient: Socrates refuses to call rhetoric a craft and classes it instead as a knack acquired by experience, a form of flattery whose aim is to gratify the audience rather than to benefit them. Its practitioner produces belief without knowledge in the souls of his hearers, standing to real expertise as pastry-cooking stands to medicine. This is an ancestor of the modern critique of spin rather than an ancient description of a named error: Plato's target is persuasive speech in general, not the specific move of reframing an unfavourable record. — Plato, Gorgias 462b–465e

Built from: Logically Fallacious — Spin Doctoring · Wikipedia — Spin (propaganda) · Stanford Encyclopedia of Philosophy — Plato's Rhetoric · The Fallacy Files — One-Sidedness

Spiritual Fallacy

Saying something plainly, and then — once it turns out not to be so — announcing you always meant it in a deeper, invisible way.

Renee sells a herbal blend to her friends and tells Kim, straight out, that it will fix her insomnia in two weeks. Kim drinks it every night and brings her sleep-tracker printout to the next get-together: no change, twelve nights of the same broken sleep. Renee glances at the printout and says, warmly, "Your body is resting on a level that watch can't measure. The healing is happening underneath." Kim came in with a plain promise and left with one that nothing could ever check.

The tell: You brought a result and the goal quietly changed shape while you were holding it. The new version sounds generous and hard to argue with, and that is exactly what bothers you — you realize there is now nothing you could have shown up with that would have counted. You feel less like you were answered and more like you were absorbed.

The move: A claim C is advanced in a literal, checkable register — a prediction with a date, a historical event, a physical effect on a body. Evidence E arrives that tells against C. The arguer then re-reads C as a 'spiritual', 'inner', or 'symbolic' claim C*, whose satisfaction conditions no observation could ever settle, and treats C* as though it had been the claim all along. E is thereby declared to have missed the point rather than to have counted against anything, and the original literal commitment is retained in the audience's memory while quietly withdrawn from testing.

Why it fails: The re-reading is introduced for one reason — the arrival of disconfirming evidence — and it is that timing, not the content of the spiritual reading, that does the damage. A claim earns credit by ruling things out; whatever C ruled out before the evidence arrived is what made believing it informative. If C can be converted, after the fact, into a claim that rules nothing out, then it never risked anything, and the confidence people placed in its literal form was purchased under false pretences. Worse, the move is unlimited: any claim whatever can be rescued this way, so the rescue is no evidence at all of the claim's merit. And the arguer usually wants both readings at once — the spiritual reading to absorb the refutation, the literal reading to keep the original authority and appeal. That is a switch between two different claims inside one sentence, dressed as a clarification.

• Was the spiritual or symbolic reading stated before the disconfirming evidence arrived, or only after it?
• Under the new reading, is there any observation at all that would count against the claim? Name one.
• Would the claim have been persuasive in the first place if it had been offered in the spiritual reading from the start?
• Is the arguer still drawing on the literal version's authority elsewhere in the same conversation, while denying it here?

Legitimate when: Texts, teachings, and promises are often figurative from the outset, and reading them that way is ordinary interpretation, not evasion — parables, liturgy, poetry, and metaphor are not literal claims that failed. The move is also legitimate as an honest revision: an arguer may say 'I was wrong to state that literally; here is the narrower thing I now think is true', provided the retraction is announced as a retraction and the new claim is not credited with the old one's track record.

Built from: Logically Fallacious — Spiritual Fallacy · Logically Fallacious — Appeal to Faith · Wikipedia — Ad hoc hypothesis · Stanford Encyclopedia of Philosophy — Fallacies

Spotlight Fallacy

Deciding what a whole group is like based on the handful of cases that made the news.

At the counter of a diner on a Saturday morning, Ray tells his granddaughter Nell that she should not let the kids walk the four blocks to school anymore. "Every time I turn the news on, some child's been snatched. It's not the world it was." Nell points out that the same three stories have been running for a week, from three different states, and that the last time anything like it happened in their county she was in high school. Ray has watched a lot of coverage; he has not seen a single number about how often it happens.

The tell: The evidence you're being given is a stack of stories rather than a rate, and every story arrives with the phrase "you hear about it all the time now." When you ask how often it actually happens, the answer comes back as another story. You get the sense the world has changed, but what changed is how much of it you're being shown.

The move: A property P is attributed to a class K on the strength of the cases of K that received media or public attention. The sample is therefore the set {k ∈ K : k was covered}, and coverage is awarded on the basis of newsworthiness — rarity, drama, harm — which is close to the inverse of frequency. The arguer treats the covered cases as a representative draw from K and infers that P holds of K generally, or that P is common in K.

Why it fails: The sampling frame is not merely small, it is systematically inverted: an outlet reports what is unusual, and an event is unusual precisely in proportion to how badly it represents the class. So the more heavily something is covered, the weaker the licence to generalise from it — the reverse of the intuition the arguer is relying on. Repetition compounds the error, because a single event carried by many outlets, and revisited over weeks, registers to a viewer as many events; ease of recall is being read as frequency, which is the availability heuristic operating on a supply of examples that was curated by someone else. The remedy is not more attentive viewing but a denominator: coverage counts events, and a claim about a class needs the rate at which they occur among all its members. Nothing in the reported cases is false; the defect lies entirely in what selected them for view.

• Is the evidence for this claim a set of reported cases rather than a rate measured over the whole group?
• Would a typical, uneventful member of this group ever have been covered at all?
• What is the base rate — how many members of the group are there, and how many of the reported kind of event occur per year among them?
• Has the number of these events actually risen, or only the number of reports about them?

Legitimate when: Coverage can be a fair sample when the covered set is close to the whole set — a small, defined group where reporting is near-exhaustive, such as every game a professional league played last season, or every airliner crash worldwide, which are counted precisely because they are rare. It is also legitimate to reason from coverage when coverage itself is the subject ('this story dominated the week'), or when an outlet supplies base rates alongside its cases so the reader can see the denominator rather than infer it.

Built from: Logically Fallacious — Spotlight Fallacy · Wikipedia — Availability heuristic · Wikipedia — Faulty generalization · Internet Encyclopedia of Philosophy — Fallacies

Statement of Conversion

Turning a sentence around — "all A are B" becomes "all B are A" — as if flipping it left it saying the same thing.

Three roommates are on the back porch after someone has been raiding the fridge. Theo lays it out: "Look, whoever's been taking my leftovers has a key to the back door." Priya nods. Then Theo says, "So it's one of us with a back-door key — that's you, me, and Nia. All three of us are suspects." Priya points out that the super has a key, the last tenant never gave hers back, and Theo's brother has one on his ring. Theo's first sentence was probably true. Turning it over made it into a different sentence entirely.

The tell: Someone says a thing you agree with, then repeats it back to you with the two halves swapped, and you nod before you notice. A beat later something itches: the second version is making a much bigger claim than the first one did, about a group nobody has actually looked at.

The move: A categorical statement relating a subject term S and a predicate term P is converted — S and P are exchanged — and the converse is asserted on the strength of the original. The move is legitimate for two of the four categorical forms and illegitimate for the other two. Illegitimately: from the universal affirmative 'All S are P' the arguer derives 'All P are S'; from the particular negative 'Some S are not P' the arguer derives 'Some P are not S'. The standard name for the error is illicit conversion; 'statement of conversion' is a schoolbook variant of the label.

Why it fails: Conversion is governed by a distribution constraint: no term may be distributed in the converse that was not distributed in the original. 'All S are P' distributes S but says nothing about all of P — it locates the S's inside the P's without claiming the P's are exhausted, so the converse asserts more about P than the original ever supplied. 'Some S are not P' distributes P but not S, and the converse would have to distribute S. The failure is therefore formal and needs no knowledge of the subject matter: 'All spaniels are dogs' is true and 'All dogs are spaniels' is false, and that single pair settles the pattern for every instance of it. In ordinary speech the error hides well, because a true universal affirmative and its false converse are both fluent English sentences about the same two words, and 'all A are B' is often heard as expressing a symmetric association rather than a one-way containment.

• Which of the four forms is being turned around — 'all ... are', 'no ... are', 'some ... are', or 'some ... are not'?
• Is the turned-around version being asserted as universal ('all P are S') on the strength of an original that was universal about S only?
• Can you name even one member of the predicate class that is not a member of the subject class? One such case refutes the converted 'all' outright.
• If the original is a 'some ... are not' statement, is the same negative being flipped? That flip never carries, whatever the terms.

Legitimate when: Two of the four forms convert freely and correctly. The universal negative converts simply: from 'No S are P' you may assert 'No P are S'. The particular affirmative converts simply: from 'Some S are P' you may assert 'Some P are S'. The universal affirmative converts only in weakened form, per accidens: from 'All S are P', assuming there are S's, you may assert 'Some P are S' — 'All spaniels are dogs' does license 'Some dogs are spaniels'. Conversion of a universal affirmative in full strength is also sound whenever the two classes are independently known to be coextensive, as in a stipulated definition ('all equilateral triangles are equiangular triangles', and conversely).

Ancient: Aristotle sets out the rules of conversion at the opening of the Prior Analytics: the universal negative and the particular affirmative convert simply, so that 'no B belongs to any A' yields 'no A belongs to any B', and 'some B belongs to some A' yields the corresponding statement with the terms exchanged. The universal affirmative, he shows, converts only to a particular — 'B belongs to all A' yields no more than 'A belongs to some B'. The doctrine settles directly that converting a universal affirmative in full strength is not permitted, which is the error this record names. — Aristotle, Prior Analytics I.2

Built from: The Fallacy Files — Illicit Conversion · Wikipedia — Conversion (logic) · Stanford Encyclopedia of Philosophy — Fallacies

Stereotyping (the fallacy)

Deciding what a person is like from what you think their group is like, and stopping there.

In the hallway of a six-unit building, the landlord, Mr. Okafor, walks Dahlia through the vacant apartment and then hands the application back before she has filled it out. "You're a nursing student, right? Students wreck places. I did it for two years and then I stopped." Dahlia has eleven years of rental history, two letters from previous landlords, and a deposit ready. He does not look at any of it. The one detail he used was the word "student."

The tell: You get a decision about you that arrived before you did. Whatever you offer — the references, the record, the actual specifics — bounces off, because the question was answered by a category you happen to fall into. The giveaway is that nothing you could say would change it, which means nothing you said was ever being weighed.

The move: A group G is associated with a property P by an unqualified generalization, 'G's are P'. An individual a is identified as a member of G. The arguer concludes that a is P, and treats membership in G as settling the question about a. Two separate steps carry the weight: the formation of the generalization, which is typically built from a small number of vivid or narratively convenient cases, and its application to a, which proceeds as though the generalization were exceptionless.

Why it fails: Both steps fail, and they fail independently. The generalization is usually unquantified — 'G's are P' does not say whether that means most, many, or the three the arguer can remember — and an unquantified generalization licenses nothing about a particular case, because it never stated a rate. Then, even where a real and well-measured group difference exists, it is almost always a difference in distributions with heavy overlap, so the group rate is at best a weak prior about any individual and is superseded the moment individuating evidence appears. This is the crucial point: the person in front of you is generally a better source of evidence about themselves than their category is, and the stereotyping move consists in refusing to update on them. It also runs a quiet substitution, treating a claim about a group's average as a claim about each member — so 'more G's than non-G's are P' comes out as 'this G is P'. Finally, the belief tends to be self-confirming, since a person judged in advance is given fewer chances to supply the evidence that would overturn the judgement.

• Is the generalization quantified at all — most, half, one in ten — or is it simply 'they're like that'?
• What is the actual evidence for the generalization, and would the arguer have accepted the same quality of evidence about a group they belong to?
• Is individuating information about this particular person available and being ignored, or is group membership the only thing being used?
• If this individual turned out not to have the property, would the arguer treat it as an exception that leaves the generalization intact? If so, the generalization is not doing evidential work at all.

Legitimate when: Reasoning from group frequencies to an individual is legitimate and often unavoidable when the generalization is genuinely measured, explicitly quantified, and used as a starting probability that the arguer will revise on individual evidence — this is what actuaries, epidemiologists, and triage protocols do, and it is why a doctor asks a patient's age before ordering tests. It is also fine when the property is definitional of the group ('licensed electricians have passed the licensing exam'). What distinguishes those uses is that they carry their uncertainty with them and yield to particular facts about the person; the fallacious use presents a rate as a certainty and treats the person as closed.

Built from: Logically Fallacious — Stereotyping (the fallacy) · Wikipedia — Stereotype · Wikipedia — Faulty generalization · Internet Encyclopedia of Philosophy — Fallacies

Stolen Concept Fallacy

Leaning on an idea in order to knock down the very thing that idea is built out of.

Casper is on the front steps of the bike shop, furious. "Owning things is a story we tell ourselves — nobody actually owns anything, people just take what they can get. Case in point: some guy walked off with my bike this morning. That's my bike. He stole it." Nadia, locking up her own, points out that he made his case using the two words his case says are made up. Casper is right that something bad happened to him. He cannot say what it was without the idea he just threw out.

The tell: You hear a big sweeping denial and something in the sentence keeps working normally, as if it hadn't been included in the demolition. It feels like watching someone saw off the branch and stay in the air. Usually you can spot the borrowed word: it's the one carrying all the outrage.

The move: An arguer asserts C, which denies the legitimacy, existence, or reliability of some notion or capacity F. But C is stated using a concept G whose meaning is parasitic on F: G cannot be defined, and the sentence containing it cannot be understood, unless F stands. The denial therefore helps itself to the very thing it revokes. The label was coined by Ayn Rand and elaborated by Nathaniel Branden ('The Stolen Concept', The Objectivist Newsletter, January 1963); Rand's own formulation is 'using a concept while denying the validity of its genetic roots, i.e., of an earlier concept(s) on which it logically depends'.

Why it fails: The defect is a presupposition failure that lands on the arguer rather than on the audience. If G's content depends on F, then anyone who understands the assertion has already granted F, so the assertion cannot be both meaningful and true — either F stands, in which case C is false, or F does not, in which case C says nothing determinate. There is no consistent restatement: strip G out and replace it with vocabulary that does not presuppose F, and either the sentence loses its content or it turns out to be a much weaker claim that was never in dispute. The move is also self-defeating in a second way, since whatever evidence or reasoning was offered for C was itself expressed in the borrowed framework and so counts equally against C. One caution belongs on the record: the label originates in mid-twentieth-century Objectivist writing and has little currency in academic logic, where the same territory is covered by self-refutation and performative contradiction; and the charge is easy to over-apply, since using a word is not always endorsing a theory about the word.

• Which specific word in the statement carries the load, and what must be true for that word to mean anything?
• Does the thing that word presupposes appear on the list of things the statement is denying?
• Can the statement be restated without that word, and does anything of the original claim survive the restatement?
• Is the arguer denying the thing itself, or only a particular theory about it? If the latter, the word may be available to them after all and no borrowing has occurred.

Legitimate when: Using a concept while rejecting a theory about that concept is entirely legitimate: a person can deny that legal title tracks moral entitlement while still using the word 'theft' in its moral sense, which is how Proudhon's slogan is usually read by its defenders. Immanent critique is also legitimate and looks identical from outside — an arguer deliberately reasons inside a framework they reject in order to derive a contradiction from it, and borrowing the opponent's vocabulary is the whole method. What distinguishes the failure is that the arguer needs the borrowed notion to hold in their own voice, for their own positive claim, and cannot get it back.

Ancient: Aristotle declines to prove the principle of non-contradiction directly and instead argues that anyone who disputes it refutes himself the moment he says anything definite: to mean one thing rather than another by a word is already to rely on the principle under attack. The disputant is asked only to say something, and in saying it he grants what he denies. — Aristotle, Metaphysics IV (Gamma), 1006a — the elenctic defence of the principle of non-contradiction

Built from: Wikipedia — Stolen concept (Self-refuting ideas) · Ayn Rand Lexicon — "Stolen Concept," Fallacy of · Logically Fallacious — Stolen Concept Fallacy · Stanford Encyclopedia of Philosophy — Fallacies

Strawman Fallacy

also known as: straw man, straw person fallacy

Instead of arguing with what you actually said, someone argues with a dumber, more extreme version they made up and pretends they beat you.

In the family group chat, Grandma texts, 'Maybe we should do a smaller Thanksgiving this year, just the immediate family, so it's less chaotic.' Uncle Ray fires back, 'So you want to cancel Thanksgiving entirely and never see any of your grandkids again? Wow, Mom, that's cold.' Grandma never said anything close to that — she just wanted a smaller dinner — but now the whole chat is arguing about 'canceling Thanksgiving,' a fight nobody actually proposed having.

The tell: You get a weird jolt of 'wait, that's not what I said' — the reply is loud and confident and sounds like a win, but it's answering a version of your point that's way more extreme or way dumber than the one that actually came out of your mouth.

The move: Person A asserts position P. Person B substitutes a distorted, weakened, exaggerated, or wholly invented position P′ that superficially resembles P. Person B refutes P′. Person B (and the audience) treat the refutation of P′ as if it were a refutation of P.

Why it fails: The refutation and the target claim are not the same claim, so defeating P′ carries no logical weight against P at all. Even a flawless demolition of P′ leaves P completely untouched — the argument commits a relevance failure by attacking a proposition other than the one actually under discussion (the Stanford Encyclopedia of Philosophy classifies it as a variety of ignoratio elenchi, irrelevant conclusion). Two separate defects are usually present at once: the opponent does not actually hold P′, and even granting that P′ is false, its falsity does not transmit to P because there is no entailment or equivalence linking the two.

• Write P (what the person actually said) and P′ (what got refuted) side by side — are they the same claim, or has scope, strength, or nuance been added or removed?
• Would the original speaker recognize and accept P′ as a fair restatement of their own view, or would they say 'that's not what I said'?
• Is the version being attacked the most defensible, representative form of the position (steelmanned), or the weakest fringe version, an out-of-context fragment, or something nobody actually claimed (hollow man)?
• Does defeating P′ do any logical work against P even in the best case — is there a stated bridge showing P′ false implies P false — or does the refutation just stop at P′ and rely on the audience not noticing the swap?

Legitimate when: Restating and then criticizing a different, weaker formulation of a claim is legitimate when the arguer explicitly flags the move as addressing a related-but-distinct position ('setting aside your main claim, consider this weaker version...'), or when the paraphrase is one the opponent has actually endorsed or that is strictly entailed by their stated view, so that refuting the paraphrase really does refute the original — this is the mirror-image legitimate technique sometimes called the 'principle of charity' or steelmanning, and its violation is precisely what makes straw-manning fallacious rather than merely disagreeable.

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Straw man · Internet Encyclopedia of Philosophy — Fallacies

Style Over Substance

Believing something because it was said beautifully, or doubting it because it was said badly.

Two roofing bids are on the kitchen table. The first came in a bound folder with drone photographs, a colour-coded timeline, and a company logo embossed on the cover; the man who dropped it off was easy to talk to and remembered the dog's name. The second is two pages, printed crooked, from a contractor who mumbled through the walkthrough. Jules says, "We're going with the first one — did you see that presentation?" Ivan reads both. The crooked pages are the only ones that mention the sheathing under the north slope, which is the whole reason the roof is leaking.

The tell: You came away persuaded and can't reconstruct what persuaded you. When you try to repeat the case to someone else, all you have is a phrase and an impression of how good it sounded in the room. That gap — strong feeling, nothing to hand over — is the thing to notice.

The move: A claim C is delivered with some presentational quality Q — polish, cadence, confident bearing, a memorable turn of phrase, high production values. The arguer treats Q as evidence bearing on whether C is true, concluding that C should be accepted. The move has a mirror form of equal standing: a claim delivered with poor Q — a stammer, an ugly slide deck, an accent, a clumsy sentence — is treated as evidence that C is false or unserious.

Why it fails: Presentational quality is produced by rehearsal, budget, temperament, and talent, and none of those is connected to the truth of what is being presented. Because a polished delivery is available to true and false claims alike, seeing polish is about as likely whether the claim is true or not, and evidence that is equally likely under both outcomes is not evidence. The mirror form fails for the same reason and does additional damage, since nervousness, unfamiliarity with the room, and a second language all degrade delivery without touching accuracy, so the rule systematically discounts the people least practised at speaking and not the people most often wrong. There is a further complication the record should name: a slogan's elegance often comes from compression, and compression is achieved by dropping the qualifications on which the claim's truth depended — so the most quotable version of a claim is frequently the least defensible one. The remedy is mechanical: state the claim flatly, in your own plain words, and see whether anything supports it.

• Strip the delivery: write the claim out in one flat sentence. Is any support left for it?
• Would the same words, said badly by a nervous person, be receiving the same credit in this room?
• Is the appeal of the line coming from what it asserts, or from its rhythm, symmetry, or memorability?
• Is the polished phrasing achieving its neatness by leaving out a qualification the claim actually needs?

Legitimate when: Presentation is a legitimate criterion when presentation is the question at issue — judging a speech contest, hiring a spokesperson, evaluating a design portfolio, assessing whether a product's interface will be usable. Delivery is also weak but genuine evidence about diligence in narrow, well-understood contexts: a brief whose citations are correctly formatted, a lab notebook kept neatly, a proposal that answered every question asked. In those cases the arguer must use it as a signal about the care taken, and must remain willing to find that a careful presenter is nevertheless wrong on the merits.

Ancient: Socrates opens his defence by remarking that his accusers spoke so persuasively that he nearly forgot who he was, and yet scarcely a word they said was true. He warns the jurors that he will not speak in their polished manner, having no experience of the courts, and asks them to disregard his style of speaking entirely and attend to one thing only: whether what he says is just. — Plato, Apology 17a–18a

Built from: Logically Fallacious — Style Over Substance · Internet Encyclopedia of Philosophy — Fallacies · Wikipedia — List of fallacies · Stanford Encyclopedia of Philosophy — Fallacies

Subjectivist Fallacy

Answering a checkable fact with "well, that may be so for you, but it isn't for me."

In the break room at the garden centre, Wendy is holding a bottle of her new blood-pressure pills and a slice of grapefruit from her lunch. Priya reads the pharmacy label out loud: do not take with grapefruit or grapefruit juice. Wendy shrugs and says, "That's true for most people. It's not true for my body." Priya asks whether her doctor told her that. Wendy says no, but she has always been able to eat whatever she wants. The label describes how the drug behaves in a liver; the reply was about how Wendy feels about labels.

The tell: You bring something anyone could look up, and the reply reroutes it into a matter of personal experience where nothing can be looked up. It is said gently, often with a smile, and it leaves you with the odd job of proving that facts apply to people. Notice the moment you start arguing about whether reality is shared instead of about the thing you came in with.

The move: A asserts a claim C whose truth does not depend on anyone's attitude toward it — a measurement, a rate, a historical fact, a physical effect. B replies, in effect, 'C may be true for you, but it is not true for me,' and treats that reply as having answered C. No argument is offered that C belongs to the class of person-relative claims; the relativity is asserted by the grammar of the reply and inherited by the audience as though it had been established. The move is also catalogued as the relativist fallacy.

Why it fails: The phrase 'true for me' is doing three jobs at once and the argument depends on not distinguishing them. It can mean 'I believe it', which is a report about the speaker and leaves C exactly where it was. It can mean 'true about me', which is a different and often perfectly legitimate claim — a person really may be an exception to a general statement, and saying so is evidence, not evasion. Or it can mean that C's truth is genuinely indexed to persons, which is correct for matters of taste and comfort and incorrect for the pollutant concentration, the drug interaction, or the standings page. The fallacious use trades on the first two readings, which are innocent, to arrive at the third, which is the one that would actually block the objection. It is also self-undermining when pushed, as Socrates presses Protagoras: if truth is whatever it is for each person, then the opponent's rejection of that doctrine is true for them, and the doctrine consumes itself. Practically, the effect is to end an inquiry that had a checkable answer available.

• Is the disputed claim the sort of thing a measurement or record could settle, or is it a matter of taste, comfort, or preference?
• Does the reply mean 'I believe otherwise', 'I am an exception', or 'the facts differ by person'? Ask which — the three have very different consequences.
• If the reply is 'I am an exception', what evidence supports that this particular person is one?
• Would the arguer accept the same reply from someone else on a question where they hold the objective view — say, the balance of their bank account?

Legitimate when: Whole classes of claims really are person-relative and the reply is simply correct for them: whether a dish is too spicy, whether a room is cold, whether a film was worth the ticket. It is equally legitimate as a claim about oneself rather than for oneself — a person who genuinely metabolises a drug differently, or whose weight is unusual for reasons of build rather than health, is reporting a fact about their own case and inviting it to be checked. Claims indexed to a stated frame are also fine: a rule that differs by jurisdiction, a deadline that differs by team, a price that differs by market. The test is whether the speaker offers something that could be examined, rather than using the qualifier to place the matter beyond examination.

Ancient: Socrates examines Protagoras' doctrine that man is the measure of all things, so that things are for each person as they appear to that person. He observes that most people believe the doctrine false; on Protagoras' own account, that belief is true for them. The doctrine therefore concedes its own falsity as often as it is denied — the reply that made every objection harmless has made itself harmless too. — Plato, Theaetetus 170e–171c

Built from: Wikipedia — Relativist fallacy · Logically Fallacious — Subjectivist Fallacy · Internet Encyclopedia of Philosophy — Fallacies · Stanford Encyclopedia of Philosophy — Fallacies

Subverted Support

Explaining why something is happening when it isn't actually happening.

At the quarterly meeting in the clubhouse, Gordon takes the floor with a printout and spends eleven minutes on why package theft has exploded on their street: the new apartments, the delivery apps, people working from home less. Two neighbours argue with him about which cause matters most. Nobody asks the other question until Rina, who keeps the incident log, says there have been three reported thefts in five years and none since March. The whole argument was about the reasons for something that was not going on.

The tell: You find yourself deep in an argument about causes and you cannot remember anyone establishing the thing being caused. It sneaks past because the detail is convincing — the more thorough the account, the more settled the underlying claim feels. The itch usually arrives as: wait, is that even true?

The move: An arguer offers an explanation E for a phenomenon Φ — 'the reason Φ happens is E'. Asserting E in that form presupposes that Φ occurs. In fact Φ does not occur, or there is no evidence that it does, and no support for Φ has been given anywhere in the argument. The audience's attention is directed to whether E is a good explanation, and in taking up that question they have already granted Φ.

Why it fails: An explanation does not argue for its own subject matter; it inherits it as a standing assumption. That is normally harmless, because the thing being explained is usually agreed on before anyone asks why it happens. Here the assumption is smuggled in behind a question about causes, and the smuggling works because objecting to an explanation feels like the natural response, while objecting to the existence of the thing explained feels like changing the subject. So the false claim travels under cover of a debate the arguer is happy to have. The failure is one of unsupported presupposition rather than of relevance or of weak evidence: the causal story may be perfectly reasonable, and the argument still supports nothing, because the load-bearing claim is the one nobody was invited to examine. Detail makes it worse rather than better — an elaborate, well-sourced account of mechanisms adds no support at all to the assertion that there is anything to account for, but it powerfully increases the audience's confidence that there is. Note the label's provenance honestly: it is a minor one, circulating through a handful of critical-thinking lists rather than through the mainstream literature.

• State the thing being explained as a flat claim of its own. Is there any evidence offered for it anywhere in the argument?
• Does the phenomenon actually occur, on the best available records — and if it does, is it occurring in the direction claimed, rising rather than falling?
• Is the discussion entirely about causes, with the occurrence of the effect never once put to the audience as something to accept or reject?
• If the phenomenon turned out not to be happening, would anything else in the argument still stand?

Legitimate when: Explaining something is entirely legitimate once its occurrence is established first, or drawn from a source the audience can check — and most explanation in science, history, and ordinary life works exactly this way. It is equally legitimate to explain conditionally, with the condition stated out loud: 'if the numbers really are rising, here is what would account for it' keeps the phenomenon on the table as a live question rather than settling it by assumption. Exploratory reasoning about a suspected pattern is also fine, provided the arguer marks the pattern as suspected.

Built from: Wikipedia — Subverted support · Logically Fallacious — Subverted Support · Internet Encyclopedia of Philosophy — Fallacies

Sunk-Cost Fallacy

also known as: Concorde fallacy, sunk cost effect

Sticking with something because of what you already put into it, even though that money or time is gone either way.

Marcus is in the waiting room at the transmission shop with a folder of receipts: eight months, four visits, twenty-one hundred dollars into a 2011 hatchback that has now thrown a code the shop wants fourteen hundred more to fix. His sister Bea, on speakerphone, asks what the car is worth. About two thousand, running. "I've got twenty-one hundred in this thing," Marcus says. "I'm not walking away now." The receipts in the folder are the same whether he pays or not. The only live question is what fourteen hundred dollars buys him today.

The tell: Every reason you can give for keeping going points backwards. You catch yourself saying "I've come too far" instead of "this is worth it," and there's a specific flavour of dread attached to stopping that is closer to embarrassment than to loss. If you imagine someone handing you this exact situation fresh this morning and you'd decline it instantly, that gap is the whole thing.

The move: A decision is faced at time t between continuing an undertaking and abandoning it. The quantities that bear on that decision are the remaining cost R and the expected benefit B of completion, both dated from t forward, compared against the best alternative use of R. The arguer instead cites S — the resources already committed and no longer recoverable — as a reason to continue, usually through the intermediate step that stopping would render S 'wasted'. S then appears in the justification even though no action now available can alter it. The aliases 'Concorde fallacy' and 'sunk cost effect' name the same pattern.

Why it fails: The decisive point is that S is identical under both options. Whether the council finishes the stadium or stops today, the eighty million is gone; a quantity that takes the same value on every branch cannot distinguish between branches, and so it can carry no weight in choosing among them. The 'waste' framing conceals this by treating waste as something the future decision determines, when in fact whether S was well spent was settled by the past — continuing does not recover a cent of it, it only adds R to the total. What the arguer is actually protecting is the standing of the earlier decision, and the empirical literature bears this out: Arkes and Blumer, in 'The psychology of sunk cost' (Organizational Behavior and Human Decision Processes 35, 1985, pp. 124–140), randomly discounted season tickets at the Ohio University Theater and found that patrons who had paid full price attended more plays over the following six months than those who received the deepest discounts, with attendance falling as the discount rose; the identified driver was a desire not to appear wasteful. The tickets bought the same plays for everyone. Only the sum already paid differed, and it should have made no difference at all.

• Is the cited cost recoverable by continuing, in whole or in part — or is it gone under every option on the table?
• Would the same choice be made if a stranger, a previous owner, or a predecessor in the job had paid that cost instead of the person deciding?
• Looking only forward from today, does the expected benefit of finishing exceed the costs still to be paid — and does it beat spending the same money on something else?
• Does the case rest on the words 'already', 'wasted', or 'come this far' where a forward-looking figure ought to be?

Legitimate when: Past investment legitimately bears on a present choice whenever it has changed the future picture rather than merely the ledger. It may have produced learning, tooling, or partial assets that genuinely lower R, so that finishing is now cheaper than starting elsewhere. It may have created contractual duties, promises, or reputational stakes with real downstream consequences — a firm known for abandoning half-finished work pays for that later, and honouring commitments has forward value. The earlier decision may also be evidence: the people who committed the money may have had information you lack, and their willingness to spend is a weak signal about the project's quality. In every one of these cases what does the work is a consequence that lies ahead, and the test is whether the argument would survive if the past expenditure had been made by someone else.

Built from: Wikipedia — Sunk cost · Logically Fallacious — Sunk Cost Fallacy · Wikipedia — List of fallacies · Stanford Encyclopedia of Philosophy — Fallacies

Suppressed Correlative

Stretching a word until its opposite has nothing left in it, then going on using the word as if the difference still meant something.

On the loading dock behind the store, Angie is telling Devon about a coworker who quietly covered a stranger's surgery bill and asked the hospital not to say who. Devon shakes his head. "Everybody's selfish. She did it because it feels good to do that. Nobody does anything they don't want to do." Angie asks what an unselfish act would look like, then. Devon thinks about it and says there isn't one — which is exactly the problem, because he is still using the word like an accusation.

The tell: A word you both understood five minutes ago has quietly grown to cover everything, and you notice you can no longer describe the difference you were trying to describe. Ask for an example of the opposite and you get a long pause. The odd part is that the word keeps its sting even after it stops meaning anything.

The move: Two terms T and T′ stand as correlatives: they partition a domain between them, and each gets its content from what the other excludes — wasteful and prudent, selfish and generous, fast and slow, coerced and voluntary. Challenged on a claim involving T, the arguer redefines T so broadly (or T′ so narrowly) that nothing whatever can fall under T′ any longer. The partition collapses; T now applies to everything in the domain. The arguer nonetheless goes on using T with the force it had when the contrast was live.

Why it fails: A contrastive term carries information exactly in proportion to what it rules out. Once every act counts as selfish, calling an act selfish tells you nothing about it — the word has been converted from a description into a label that attaches automatically, and no observation could now bear on its application. That much would merely be an unhelpful definition. The failure comes from what the arguer does next: the emotive and evaluative weight of T was earned under the old, contrastive meaning, and it is that weight the argument spends. So the move wins a definitional point in the broad sense and cashes it in the narrow sense, which is an equivocation across two meanings of one word inside a single argument. Nothing about the world has changed, either — the distinction the original dispute was about has not been refuted, only evicted from its vocabulary, and it can be restated immediately in fresh words: if all spending is now an expression of priorities, we still need terms for spending that serves its stated purpose and spending that does not. The Scottish logician Alexander Bain treated the pattern in the nineteenth century among what he called fallacies of relativity.

• Which pair of opposed terms is the argument built on, and what did each of them exclude before the redefinition?
• Under the offered definition, could anything at all still fall on the other side? Ask for one example — the inability to supply one is the diagnosis.
• Is the redefined term still being used with the punch it had under the old meaning, as praise, blame, or a warning?
• Can the original disagreement be restated using different words? If it can, the redefinition changed the vocabulary and not the question.

Legitimate when: It is legitimate — and often important — to reject a false either/or, to argue that a term admits of degrees rather than a clean cut, or to propose a technical redefinition openly for a specific purpose, as economists do with 'rational' and biologists with 'fitness'. What keeps those moves honest is that they supply a replacement contrast: a redefinition that draws the line somewhere new is doing work, while one that erases the line is not. Saying 'I'm somewhat hungry' rejects the offered pair without dissolving it; saying 'everyone is always hungry, since anyone could eat' dissolves it. The other honest form is an empirical claim that one side of the pair happens to be empty — that no case of T′ has ever been found — which is a substantive claim, open to counterexample, rather than a stipulation immune to one.

Built from: Wikipedia — Suppressed correlative · Logically Fallacious — Suppressed Correlative · Wikipedia — List of fallacies · Stanford Encyclopedia of Philosophy — Fallacies

Survivorship Fallacy

also known as: survivorship bias

Counting only the ones that made it, and then deciding what made them special.

At an estate sale in a two-car garage, Warren runs his hand along a maple dresser and tells his nephew what he tells him every time: "Nineteen fifties. They built things to last back then — look at this joinery. Nothing you buy today survives ten years." The nephew, who has been to a lot of these with him, points out that they are standing in a garage full of things that lasted seventy years, which is why they are here to be looked at. Everything cheap and badly made from 1954 went to the dump decades ago. The good stuff is all that's left to see.

The tell: Every piece of evidence you're being shown is a success story, and when you try to think where you'd go to find the ones that failed, you realize there's nowhere to look — they left no trace. Advice built this way has a particular sound: it's stated like a rule, but its whole backing is a hall of fame.

The move: A property P is estimated for a class K from the members of K that are available to be observed. Availability is conditional on having passed some filter F — surviving, staying in business, remaining standing, still being talked about — and cases that failed F are absent from the sample not by chance but by construction. The arguer then reads the rate of P among the observed cases as the rate of P in K, or reads whatever the observed cases have in common as the cause of passing F. The label also travels as survivorship bias.

Why it fails: The sample has been defined by the very outcome under study, so that outcome's rate cannot be recovered from it: every case in front of the arguer passed the filter, which is a fact about how the sample was assembled and not a fact about the process. Two distinct errors follow. The rate is inflated, because the cases that would have lowered it were removed from the denominator before anyone counted. And causal attribution is corrupted, because any feature shared by the survivors will look like the reason they survived even if the non-survivors shared it equally — there is no comparison group left against which to check. What makes this different in kind from ordinary small-sample error is that collecting more data does not help: every additional case arrives through the same filter, so the bias does not shrink as the sample grows, and diligence in observation makes the illusion sharper rather than weaker. The repair is never more observations of the same kind. It is either restoring the missing cases to the denominator from records, or reasoning indirectly about what their absence implies — which is precisely what Abraham Wald did with the bombers.

• What did a case have to do to end up in front of you — succeed, survive, stay open, stay visible?
• Where are the cases that failed that test, and could they be counted even in principle: closed funds, demolished buildings, bands that never got signed, patients who did not return?
• Do the missing cases share the feature being credited for the survivors' success? If they do, the feature explains nothing.
• Restore the failures to the denominator and recompute: what is the rate now, and is it still remarkable?

Legitimate when: Conditioning on survival is harmless when the filter is independent of the property being studied — if what determined whether a case reached you has nothing to do with the question you are asking, the sample is fine. It is also legitimate whenever the non-survivors are recovered and counted rather than assumed away: fund databases that retain dead funds, clinical trials analysed by intention to treat, and Wald's own method of estimating vulnerability from the damage patterns of survivors are all serious statistical work built on exactly this awareness. And the question itself may properly be about survivors: 'what do the buildings that lasted four hundred years have in common?' is a good question, so long as the answer is not then sold as a recipe, since the buildings that shared those features and fell are not in the photograph.

Ancient: Cicero has the character Cotta recount an anecdote about Diagoras of Melos, called the Atheist. Shown a temple hung with votive tablets dedicated by sailors who had prayed in a storm and come safely to shore, Diagoras was asked whether this did not prove the gods heed prayer. He answered by asking where the tablets were of all those who prayed just as hard and drowned. The display could only ever contain survivors, and its persuasive force came entirely from that fact. — Cicero, De Natura Deorum III.89

Built from: Wikipedia — Survivorship bias · Wikipedia — Abraham Wald · Wikipedia — Faulty generalization · The Fallacy Files — Hasty Generalization

Texas Sharpshooter Fallacy

also known as: clustering illusion

Firing first and painting the target around the holes afterward, then calling yourself a great shot.

Five people in a thirty-person office come down with the same stomach bug in one week. Kevin prints the seating chart, circles four of the five desks along the north wall, and announces at the Monday standup that it's the old window unit blowing spores. He is very convincing about it. Nobody circled the twenty other people along that same wall who are fine, and nobody said before the outbreak that the north wall was the thing to watch. Kevin drew the circle after he knew who was sick.

The tell: The pattern always arrives already explained, and the boundary around it is oddly specific — these three months, this side of the building, this age range. When you ask about a case just outside the line, the line moves to include it or the case is set aside as unrelated. That flexibility is the giveaway: a real edge doesn't slide.

The move: A body of data D is examined. A subregion or subset R of D — a run of months, a block of streets, a slice of a population — is then singled out because it displays a striking pattern. A hypothesis H, that some cause is responsible for the pattern in R, is formulated with R already in view. The fit of H to D restricted to R is then offered as confirmation of H. The boundary of R is a free parameter, fixed after the outcomes were seen, which is what the name records: the marksman fires at the barn and paints the target around the tightest cluster of holes.

Why it fails: The evidential weight of a match depends entirely on how unlikely the match was in advance, and when the target is drawn after the shots there is no 'in advance' to speak of. The number of ways any noisy dataset can be carved into some striking-looking subset is very large, so finding one such subset is not merely possible under pure chance but expected — R. D. Clarke's 1946 analysis of wartime flying-bomb impacts on London found that the apparent targeting patterns Londoners were sure they could see fitted a Poisson distribution, which is to say randomness. The arguer therefore reports the wrong probability: the honest figure is not how unlikely a cluster this size was at that spot, but how likely some such cluster was somewhere in data of this size and noisiness, and that second number is usually unremarkable. The Swedish power-line study of 1993 is the canonical illustration — researchers examined more than eight hundred medical conditions among people living near high-voltage lines and found a fourfold childhood leukaemia rate; with that many conditions inspected, some result was almost certain to reach significance, and the association did not hold up in later work. What is fatal is not noticing the cluster; it is counting the noticing as the test.

• Was the specific pattern — which months, which streets, which subgroup — written down before the data were examined?
• How many other regions or subsets could have been circled instead, had they been the ones that happened to look striking?
• Does the effect hold outside the boundary that was drawn around it, or does it live entirely inside that line?
• Has it reappeared in a fresh, independently collected dataset in which the target was fixed in advance?

Legitimate when: Noticing a cluster and forming a hypothesis from it is how a great deal of discovery works — John Snow's map of cholera deaths around the Broad Street pump is the standing example, and epidemiology would be impossible without such looking. The move is legitimate when the post hoc cluster is treated as generating a hypothesis and the hypothesis is then tested against data that did not generate it, or when the size of the search is disclosed and corrected for. It is also fine when the target genuinely was specified in advance: a prediction registered before the season, a region flagged on prior mechanistic grounds, an effect the theory required before anyone measured it.

Built from: Wikipedia — Texas sharpshooter fallacy · The Fallacy Files — The Texas Sharpshooter Fallacy · Wikipedia — Clustering illusion · Internet Encyclopedia of Philosophy — Fallacies

Tokenism

Pointing at one small gesture as proof that a much bigger problem has been handled.

At parent night in the school library, three families raise the same thing they raised in the spring: a group of eighth graders who have been making one boy's life miserable since September, and nothing changing. Principal Vance says, with real warmth, "We held the kindness assembly in October — the whole grade attended, and we brought in a speaker." A mother asks what happened after October. He mentions the assembly again. It was one afternoon, nine months ago, and it is the entire answer.

The tell: You asked about a pattern and you got one story, delivered with a certain relief, as though producing it ended the meeting. Ask a second time and you get the same story. The feeling is of a door closing gently: nothing was refused, but nothing else is coming.

The move: A standard, obligation, or complaint O concerns practice at scale — how an institution generally hires, serves, accommodates, or behaves. Challenged on O, the arguer cites a single instance G that visibly satisfies some small fraction of O, and concludes that O has been met and the question is closed. Two steps carry the argument: an inference from one instance to a general practice, and a substitution of a symbolic act for a substantive one. As an argument pattern this is distinct from the social practice the word usually names, though it is what the practice is designed to make available.

Why it fails: The sample is a single case, and it is the worst possible single case, because it was not drawn from practice — it was produced in order to be pointed at. An instance selected for its presentability carries no information about typical practice, in the way that a showroom carries no information about a factory floor. Beneath that lies a quantity question the argument depends on never being asked: the gesture discharges some measurable fraction of the obligation, and naming that fraction usually settles the matter outright — one liaison against a service gap covering forty counties, one accessible entrance out of nine. The move also reverses the burden. What was requested was evidence about the whole; what is supplied is an existence proof of one exception, and an existence proof cannot answer a question about a distribution. Finally, it trades on the sincerity of the gesture, which is generally real: the single hire is a genuine hire, the single change genuinely happened, and it is the reality of the instance that makes the leap to sufficiency feel less like an argument than like a fact.

• What proportion of the stated problem does the cited action actually address — can anyone put a number to it?
• Is this the only instance the speaker can name, and was it produced in response to the complaint itself?
• Would practice at scale look measurably different if the gesture had never happened?
• When asked to show the general pattern rather than the one example, does the speaker produce it or repeat the example?

Legitimate when: A single instance answers a question about scale when the obligation genuinely was singular — one ramp was the whole requirement, one signature was what the rule asked for — or when the instance is offered as a randomly drawn or representative sample of practice rather than as its showpiece. It is also entirely legitimate as a first step honestly labelled: 'this is one hire, it addresses maybe a tenth of the gap, and here is the schedule for the rest' concedes the scale question rather than closing it, and that concession is the difference. Symbolic acts can also be defended on their own terms — a gesture may matter as a signal of intent even when nobody claims it solves anything — provided the claim being made for it is the symbolic one.

Built from: Logically Fallacious — Tokenism · Wikipedia — Tokenism · Internet Encyclopedia of Philosophy — Fallacies · Wikipedia — Faulty generalization

Traitorous Critic Fallacy

Answering a complaint by questioning whose side you're really on and suggesting you go somewhere else.

On the running club's message board, Ines posts photographs of the Saturday long-run route: two unlit miles along a shoulder with no verge, and a blind curve where a car came within a foot of the lead group in March. She proposes moving it a half mile east. The first reply, from a board member of eleven years, reads: "Some of us love this club. If it's really that bad, the trail runners over in Ashford are always taking new members." Nineteen replies later, the thread is about whether Ines is a real club member. The blind curve has not come up again.

The tell: You brought a specific problem and within two exchanges the subject is you — your commitment, your attitude, how long you've been around. It is often said warmly, which is disarming, and you notice you have started listing the reasons you care about the place instead of the reasons the thing you raised is a problem.

The move: A critic C, who belongs to a group G, advances a criticism K about G — a practice, a rule, a condition. Rather than engaging K, the respondent attributes C's criticism to divided loyalty or to an attachment to some out-group, and invites C to depart: if you dislike it here so much, leave. The reply concerns C's standing, affections, and available exits, and leaves the content of K entirely untouched. It circulates in the literature under the Latin tag argumentum ergo decedo.

Why it fails: Whether the race route is unsafe, the staffing ratio unlawful, or the rule inconsistently applied is settled by facts about routes, staffing, and rules; the critic's affections and the availability of somewhere else to go bear on none of them. That much is a plain failure of relevance, but the move does two further things worth recording. It converts a first-order question into a loyalty test, so that any reply the critic makes — including a denial of disloyalty — concedes the new frame and abandons K. And it is self-sealing: because criticism is itself treated as the evidence of divided loyalty, every possible critic is disqualified by the act of criticising, which leaves the group structurally unable to receive correction from inside. It also gets the direction of the evidence backwards. Someone who stays and raises an unpopular problem, at cost to their standing, has if anything demonstrated more attachment than someone who leaves quietly; exit is the cheap option, and the reply treats it as the loyal one.

• Has any part of the reply addressed what the criticism actually claims, or only the critic's standing and options?
• Would the criticism be true or false regardless of whether the critic stayed, left, or never belonged in the first place?
• Is any independent evidence offered for the alleged outside allegiance, or is the criticism itself being treated as the proof of it?
• Is there any way to raise this objection that would not attract the same reply? If not, the reply is not responding to anything about this objection.

Legitimate when: Affiliation is a legitimate topic when standing genuinely is the question at issue: a conflict-of-interest disclosure, whether someone should hold an office or a vote, whether a group's membership requirements were met, whether a witness has an undisclosed stake. Advice to leave can also be honest and kind when it is offered as advice about a person's wellbeing rather than as a reply — 'you have been miserable here for two years and might be happier elsewhere' is a real thing to say, as long as it is marked as separate from the merits and the merits are addressed too. And where someone has a documented record of preferring and promoting a rival group while disparaging this one, that record is admissible evidence about their reliability as a reporter, though never about whether their specific claim is correct.

Built from: Wikipedia — Ergo decedo · Logically Fallacious — Traitorous Critic Fallacy · The Fallacy Files — Argumentum ad Hominem · Wikipedia — List of fallacies

Two Wrongs Make a Right

Saying it's fine to do something wrong because somebody else did it first.

Tasha comes home late to find a silver sedan in her assigned parking space, again, and Rob's apartment light on. She knocks. Rob opens the door already talking: "In February you blocked my truck in for two days when your sister visited. Two days, Tasha." She had, and she had apologized for it, and it is also eleven at night and there is nowhere else to park. Every sentence Rob says is about February. None of them are about where she is supposed to put her car tonight.

The tell: You raise one specific thing and the reply opens with "well, when you —" and suddenly you're litigating an incident from months ago. You can feel the conversation turn into a ledger. The tell is that you're now defending yourself about something you weren't there to discuss, and the original thing has quietly gone unanswered.

The move: An act A performed by the arguer, or by the arguer's side, is challenged as wrong under some standard S. The arguer cites a prior wrong B committed by another party — often the challenger, often the challenger's side — and concludes that A is permissible, or that no one is now in a position to object to it. The reason offered is that B occurred. Nothing is said to show that S does not apply to A, that S is mistaken, or that A satisfies S.

Why it fails: A standard governing conduct is not a running score between parties. That someone else violated S is not a way of complying with S, and another party's breach does not repeal it — otherwise the standard would apply only to whoever moved first, which no one actually believes. The argument also runs together two questions that come apart cleanly: whether A is wrong, and whether this particular objector has the standing to say so. Even where the second is fully conceded — the challenger may be a hypocrite of long standing — the first is untouched, because a person can be right about a rule they themselves break. Generalised into a policy the move is worse than merely unsupported: accepted as a rule, 'they did it first' licenses every party in a dispute to match and exceed the last offence, which is the escalation the proverb exists to name. The honest neighbours are the reason the move feels plausible. Self-defence, restitution, proportionate retaliation, and deterrent sanction are all governed by their own standards, which specify who may respond, how much, aimed at what end; the two-wrongs move borrows their moral weight while supplying none of their conditions.

• Does the prior wrong change the standard that applies to this act, or only the temperature of the room?
• Is the response bounded — proportionate, aimed at repair, restitution, or deterrence — or does it simply repeat the original wrong at the same or greater scale?
• Would the arguer accept 'you did it too' as a complete defence if the roles were reversed?
• Strike the prior wrong from the record: is there anything left that defends the act on its own terms?

Legitimate when: Several legitimate moves sit close by. Self-defence and proportionate retaliation genuinely are governed by what the other party did, but they carry conditions — a threat that is present, a response scaled to it, an end beyond the harm itself. Restitution likewise: taking back your own property from the person who took it is answerable to the fact that it is yours, not to their having wronged you. Arguments from consistency and fair notice are also real: pointing out that a rule has gone unenforced for years is a substantive claim about the rule's standing and about selective enforcement, and it belongs in the discussion. What separates all of these from the fallacy is that each argues about the standard — its content, its scope, its application — rather than treating the prior wrong as a licence that suspends it.

Ancient: Socrates gets Crito to agree, step by careful step, that one must never do injustice — and then that one must not do injustice in return for injustice suffered, nor do harm in return for harm, however one has been treated. He notes that few people accept this, and that those who do and those who do not have no common ground for discussion. It is an explicit ancient rejection of the principle that a wrong received licenses a wrong returned. — Plato, Crito 49a–e

Built from: Wikipedia — Two wrongs make a right · Logically Fallacious — Two Wrongs Make a Right · Internet Encyclopedia of Philosophy — Fallacies · Stanford Encyclopedia of Philosophy — Fallacies

Type-Token Fallacy

also known as: type-token confusion

Mixing up a kind of thing with the actual physical things of that kind — 'the same jacket' meaning the same design, versus meaning the one coat you both share.

At a record shop on a Saturday, Priya tells her brother Wes she just bought the same album he has. Wes goes quiet and says, "You went in my room and took it?" Priya stares at him. "No. I bought a copy. It's in the bag." Wes had heard "the same album" as the same physical record standing on his shelf; Priya meant the same release, one of thousands pressed. They spent ten minutes arguing about a theft that had not happened, using the same four words the whole time.

The tell: You get accused of something that would have been impossible, or you find yourself defending a number that isn't wrong. There is a stubborn flatness to it — the other person keeps repeating a word you also used, in a tone that says the matter is closed, and you cannot locate the disagreement because you agree on every word in the sentence.

The move: A count noun T admits two counting criteria. Under the type criterion it counts repeatable kinds — designs, models, titles, patterns, word-forms. Under the token criterion it counts the individual concrete instances that realize those kinds. One step of the argument fixes T under one criterion and another step fixes it under the other, and a count, a property, or an identity claim is then carried across the two as though a single subject had been in play throughout. The characteristic surface signature is either a numerical mismatch presented as a contradiction, or a claim of sameness ('the same X') that is true of the shared type and false of any two tokens, or the reverse.

Why it fails: Types and tokens are not two names for one subject; they are two things standing in the instantiation relation, and the predicates that attach to each are largely disjoint. A type can be popular, discontinued, adopted by a school board, or protected by copyright; a token can be scratched, sitting in a garage, sold on Tuesday, or lost. When one step counts or predicates of the type and the next helps itself to the token, the noun that appeared to hold the two steps together has silently changed what it picks out, so the steps are not about the same thing and the second is not supported by the first. Nothing has been established about a type by establishing it of one token, because a token may be defective while the type is not; and nothing has been established about a token by establishing it of the type, because the type is a pattern that no single instance need fully carry. What looks like an argument is therefore running on the grammar of a shared count noun rather than on any relation between the two subjects, which is why the mistake so often surfaces as an accusation of lying: two perfectly correct numbers are set against each other as if they had counted the same things.

• Take each occurrence of the disputed noun and ask of it: is this counting kinds, or counting individual physical instances?
• Rewrite the passage with 'kind of X' at one step and 'this particular X' at the other — does the argument still hold together, or does it fall into two unrelated claims?
• Is the property being carried across one that can hold of a design or pattern (popular, discontinued, adopted, patented), or only of a physical instance (scratched, missing, shipped, sold)?
• If two figures are being treated as a contradiction, are they counting the same things — or is one counting titles and the other counting copies?

Legitimate when: Traffic between the two levels is entirely legitimate when a convention fixes it and both parties honour that convention. 'The Corolla has sold fifty million units' predicates a token-count of a type by an understood idiom, and 'we own the same dishwasher' is an ordinary true claim about a shared model that no one mistakes for co-ownership. The move goes wrong only when one step relies on the type reading and another relies on the token reading while the argument needs one subject running all the way through.

Built from: Stanford Encyclopedia of Philosophy — Types and Tokens · Wikipedia — Type–token distinction · Logically Fallacious — Type-Token Fallacy · Stanford Encyclopedia of Philosophy — Fallacies

Unfalsifiability

Holding a belief in a shape where nothing that could ever happen would count against it, and then pointing at how it has never once been wrong.

Marcus tells his housemates that the landlord is deliberately messing with them. The heat cuts out in November — see, that's him. The heat is fixed by that afternoon — "right, because he knows we're onto him, so now he's covering." The building manager sends a friendly note about the boiler schedule — "that's the tell, nobody is that nice." Every single thing the landlord could possibly do has already been assigned a meaning in advance, and all the meanings point the same direction.

The tell: You keep offering things and none of them land. Whatever you bring — good news, bad news, a receipt, a photo — gets folded straight back in as more support. After a while you can predict their answer before you finish your sentence, and that is the moment to stop: there is nothing you could possibly say that would move them, which means nothing they said was ever really being tested.

The move: A claim C is advanced as though it reported something about the world, but it is held in a form under which no possible observation is admitted as counting against it: every outcome, including outcomes that look contrary, is absorbed as further confirmation, frequently by an auxiliary condition attached after the awkward result arrives. The arguer then cites C's unbroken record of confirmation as support for C, and reasons onward to practical claims that only a genuinely empirical thesis could license. Because the class of observations C excludes is empty, C divides the possibilities nowhere.

Why it fails: Evidential support is comparative: an observation supports a claim only to the extent that it was more to be expected if the claim were true than if it were false, which requires the claim to forbid something. A statement compatible with every outcome forbids nothing, so no outcome discriminates between its truth and its falsity, and its perfect record of confirmation carries about as much information as a two-headed coin carries about luck. The defect is worse when the immunity is manufactured retroactively — each time a counter-instance appears, an unmotivated condition is added to swallow it, so the claim survives not by matching the world but by being reshaped to fit whatever the world does. What is offered as an empirical thesis with overwhelming support turns out, in the form it is actually held, not to be an empirical thesis at all, while continuing to be spent as though it were. The rhetorical damage is symmetrical: the arguer cannot lose the argument, and for exactly that reason has not won it.

• Ask the arguer directly: what specific thing, if you observed it, would you accept as counting against this? If no answer is forthcoming, the mark is present.
• Are an outcome and its opposite both being read as support for the same claim?
• When a counter-instance was raised, was a new condition added to the claim afterwards — and is there any reason to believe that condition other than that it saves the claim?
• Does the claim, as now stated, rule out any state of affairs at all? Ask for one, by name.

Legitimate when: Plenty of respectable claims are not testable by observation — mathematical results, definitions, moral principles, metaphysical theses — and holding them is no defect, provided they are not dressed as empirical findings and not credited with evidential support they cannot receive. Working science also shields a core theory with auxiliary assumptions, and that is legitimate practice; it turns into this error only when the auxiliaries are introduced solely to absorb the anomaly and carry no independent motivation and no testable consequence of their own.

Built from: Stanford Encyclopedia of Philosophy — Karl Popper · Logically Fallacious — Unfalsifiability · Stanford Encyclopedia of Philosophy — Fallacies

Unreasonable Inclusion Fallacy

Stretching a heavy word to cover something much milder, then treating the mild thing as if it deserved everything the heavy word carries.

Someone posts in the neighborhood group that the community garden has started locking the tool shed overnight. A commenter replies: "So they're stealing from the neighborhood now." Half the thread runs with it — thieves, we should call somebody, this is what happens. Nothing was taken from anyone. A shed that used to stand open at two in the morning now stands closed at two in the morning. But the word had already done its work, and by Sunday three people believed there was a theft ring in the garden.

The tell: The word lands much harder than the thing did. You find yourself agreeing that what happened was annoying, or even wrong, while wanting to say "that isn't what that word means" — and you can feel in advance that saying it will make you sound like you are defending whatever the big word usually names. That trap is the move working.

The move: A label L carries heavy condemnatory force because of the paradigm cases it was coined for — cases sharing features f₁…fₙ that account for that force. The arguer applies L to a case c that lacks most of f₁…fₙ, and then draws on the whole of the condemnation L earned from the paradigm cases in order to condemn c. The criterion under which c was admitted to L's extension is never stated or defended; the widened extension is asserted by the act of labelling, while the original charge is quietly retained.

Why it fails: A word's condemnatory weight is not free-floating. It is owed to the features of the cases that made the word necessary, and it travels with those features rather than with the letters. Widening a term's extension while holding its force constant is therefore a bookkeeping error with two possible resolutions: either the new case genuinely shares what made the paradigm cases grave, in which case that has to be shown by argument, or it does not, in which case importing the force is unearned. The move runs on presumption rather than support, because the disputed step — that c belongs under L at all — is smuggled in as a premise by the mere application of the name, so the audience finds itself reacting to c at a severity no stated reason supports. Two costs follow. The person or act named is condemned beyond anything argued for, and the word itself is spent: it loses the power to mark the cases it was built for, so the next real instance arrives with the vocabulary already worn out.

• Name the features that made the label's paradigm cases serious. Does this case actually have them?
• Has any criterion for including this case under the label been stated, or was the label simply applied and left to do its own work?
• Swap the label for a plain description of what physically happened. Does the same reaction still follow from the description alone?
• Is the label's severity being borrowed while its usual requirements are being dropped — that is, would the case fail the label's own standard test?

Legitimate when: Words legitimately widen, and sometimes they should: a term coined for one shape of harm can be correctly extended to a newly recognised case that genuinely shares what made the original cases wrong, and arguing for such an extension is respectable and often important work. The move is an error only when the extension is asserted rather than argued, and when the term's inherited force is kept for a case lacking the very features that force was tracking.

Built from: Logically Fallacious — Unreasonable Inclusion Fallacy · Wikipedia — Concept creep · Stanford Encyclopedia of Philosophy — Fallacies

Unwarranted Contrast

Hearing what someone left out as though they had said it, and then arguing against the thing they never actually said.

At the reception, the groom's brother stands up and thanks Rina's parents for everything — the car, the flowers, the long weekend of folding programs at their kitchen table. Then he sits down. Across the room, the groom's mother has gone very still. By the drive home it has hardened into fact: he thanked her side and not ours, so he thinks we did nothing. He had said nothing at all about her side. He had run out of nerve and sat down.

The tell: You are being held to a comparison you never drew. There is a moment where you want to say "I didn't say that" and you realize it will be heard as a dodge, because to the other person the second half of the sentence was every bit as loud as the first. The feeling is of being cross-examined about a line that exists only in someone else's head.

The move: A speaker asserts S. In ordinary conversation S carries a suggestion beyond what it states — characteristically a scalar suggestion, where asserting the weaker item on a scale ('some', 'tried', 'competent') hints that the stronger one does not hold, or a suggestion of contrast, where mentioning one item hints that the unmentioned alternatives lack the property. The arguer takes that suggestion to be part of what S states, and then reasons from the contrast as though it had been asserted and supported. The signature step runs from 'some X are F' to 'some X are not F', or from 'we are F' to 'they are not F', with nothing behind it but emphasis or selective mention.

Why it fails: What a sentence states and what uttering it suggests come apart, and there is a standard test that shows this: a suggestion of this kind can be cancelled without contradiction — 'some of them came; in fact, all of them did' is surprising but not self-defeating — whereas nothing a sentence genuinely states can be withdrawn that way. Such suggestions arise from expectations about how a cooperative and fully informed speaker would have talked, not from the meanings of the words, so they are defeated by anything that explains the speaker's wording otherwise: ignorance, discretion, relevance, tact, or simple economy of breath. Treating the suggestion as asserted therefore attributes to the speaker a claim they did not make, and, more damagingly, credits that claim with whatever support the speaker offered for the weaker thing they did say. Nothing in the passage bears on the contrast at all, because the contrast was assembled by the hearer out of what went unsaid.

• Is the contrast stated anywhere in the actual words, or is it carried only by emphasis, by the choice of a weaker term, or by what was left out?
• Try cancelling it: append 'in fact, the other one too' or 'in fact, all of them'. Is the result a flat contradiction, or merely a surprise? If merely a surprise, the contrast was never stated.
• Is there any evidence in the passage that the unmentioned alternative lacks the property — or only silence about it?
• Could the speaker's wording be fully explained by caution, ignorance, politeness, or relevance rather than by an intended contrast?

Legitimate when: Reading a suggested contrast is very often exactly right, and conversation would be unworkable without it: when a speaker is known to be fully informed and is speaking under an obligation to disclose everything relevant — a witness under oath, an auditor's report, a safety disclosure — the step from 'some' to 'not all' is a reasonable reading of what they meant, and can be pressed. It is legitimate whenever the contrast is drawn as a defeasible reading of the speaker's meaning and offered as such; it becomes an error when the reading is treated as something the speaker asserted and then carries the weight of an established claim.

Built from: Logically Fallacious — Unwarranted Contrast · Stanford Encyclopedia of Philosophy — Implicature · Stanford Encyclopedia of Philosophy — Fallacies

Use-Mention Error

also known as: use-mention confusion

Treating a word as though it were the thing it points at — arguing about the label when the argument was supposed to be about what the label names.

Mr. Okafor writes on the whiteboard a list headed 'Words we do not use in this room,' and one of the entries is a slur. A student's hand goes up: "You just wrote it on the board. So you used it." Half the class nods; it is right there in blue marker. But the board is a list of which words are off limits, not an instance of calling anyone one of them. What is on the board is the word being pointed at, not the word being aimed at somebody.

The tell: The argument has suddenly become about spelling. Somebody is holding up a page or a screenshot as if a word's mere presence on it settles what the page does. You want to say "that's the word, not the thing" and you can feel it will sound like a technicality, even though it is the entire issue.

The move: An expression E can occur in a passage in two ways. It is used when it does its ordinary job of picking something out, and it is mentioned when the expression itself is the thing under discussion — conventionally marked by quotation marks in philosophy or italics in linguistics. The argument attaches to what E picks out a predicate that belongs only to E-the-expression (its spelling, length, sound, its appearance in a document, its being on a list), or attaches to E-the-expression a predicate that belongs only to the thing. The two steps look like they concern one subject because the same letters occur in both, but the subjects are an expression and its bearer, which are distinct objects.

Why it fails: A name and its bearer are two things, and almost nothing true of one is true of the other: 'Chicago' has seven letters and Chicago has millions of people, and neither fact transfers. Quotation is not typographic decoration but the device that tells the reader which of the two is being talked about, so an argument that omits it has not merely been untidy — it has failed to specify its own subject, and cannot be assessed until the marks are put back. When such an argument nonetheless appears to go through, what carries it is the shared spelling, which is a fact about orthography rather than about the world. Two tests expose it. Substituting a synonym leaves the truth of a step about the thing untouched but can wreck a step about the word, so any step that turns on the exact wording was about the word. And restoring quotation typically splits the argument into two unrelated claims: one trivially true of an expression, one plainly false of what the expression names. The characteristic damage is a category error presented as a discovery — a property of words announced as a property of things, or a rule about things enforced against marks on a page.

• Put quotation marks around every occurrence where the expression itself is the topic. Does the argument still read as one connected argument, or does it break into two claims about different subjects?
• Is the property being asserted one that can hold of expressions — spelling, length, being printed somewhere, appearing on a list — or only of things in the world?
• Replace the disputed term with an exact synonym. Does the truth of that step change? If it does, that step was about the word, not the thing.
• Does the argument treat a word's appearing in a text as the thing itself being present in the world?

Legitimate when: Talking about expressions is entirely legitimate and constantly necessary: lexicography, editing, statutory drafting, translation, code review, and teaching all consist largely of mentioning expressions rather than using them. The requirement is only that the passage mark which is which and keep each predicate on its proper side. An argument may also move between the word and the thing whenever it states the bridge explicitly — as when a statute's exact wording really is what is at issue, because in law the wording is what governs.

Ancient: Among the arguments Diogenes Laertius reports Chrysippus propounding is this one: if you say something, it passes through your lips; you say 'wagon'; therefore a wagon passes through your lips. Diogenes notes that some of these puzzles were credited instead to Eubulides. The joke works precisely on the gap this record names — what passes the lips is the word, while the thing the word names stays in the yard — though the ancients had no term for the distinction, which was not articulated until modern logic; so the passage is an attested ancient instance of the confusion being exploited, not an ancient naming of it. — Diogenes Laertius, Lives of Eminent Philosophers VII.186–187 (reporting Chrysippus)

Built from: Wikipedia — Use–mention distinction · Diogenes Laertius, Lives of Eminent Philosophers VII (Perseus) · Stanford Encyclopedia of Philosophy — Quotation · Stanford Encyclopedia of Philosophy — Fallacies

Weak Analogy

Arguing that two things will behave the same because they resemble each other, when the ways they resemble each other have nothing to do with the thing being predicted.

On the sideline at an under-11 soccer game, a parent who runs a landscaping crew corners the volunteer coach. "Look, a team is a team. I've managed crews for eighteen years — you set the standard, you bench whoever doesn't meet it, and the rest fall in line. Sit the two slow kids for a game and watch the whole squad tighten up." The coach doesn't say anything. He is thinking about the fact that his crew gets paid and these kids do not, and that the slow ones are nine.

The tell: The comparison arrives already finished, usually with a little flourish — "it's exactly like" — and it feels satisfying before you've checked anything. What you notice, if you slow down, is that you agree the two things are alike and you still don't believe the point. That gap between granting the likeness and refusing the ending is the signal; the likeness is real and it isn't the likeness that matters.

The move: A source case A and a target case B are observed to share features f₁…fₙ. A is known to have a further property P. The arguer concludes that B has P, or is likely to. The argument is weak when the shared features f₁…fₙ bear no known connection — causal, structural, or explanatory — to P, when the features doing the persuasive work are merely salient rather than relevant, or when there are known differences between A and B that fall precisely where P is determined. Its strength is a matter of degree rather than a switch, and it is properly assessed against the weight the conclusion is being asked to carry.

Why it fails: Any two things resemble each other in indefinitely many respects and differ in indefinitely many others, so bare similarity supplies no support at all; what supports the projection is a reason to think the respects in which A and B are alike are the respects on which P depends. When that reason is missing, the argument is running on the vividness of the comparison rather than on any connection between the compared features and the projected one, and the audience's sense that the two cases 'feel alike' is doing the work a stated link should do. The failure is aggravated by the disanalogies: an analogical argument is only as good as its worst relevant difference, and a single difference located at the point where P is fixed can wipe out any number of similarities elsewhere. This is why the correct reply to a weak analogy is rarely 'they are not alike' — they usually are, in the ways named — but 'those are not the ways that matter for this.' Because the support is a matter of degree, the same comparison can be respectable as a heuristic that suggests a line of inquiry and disgraceful as grounds for a decision, and the arguer's error is very often one of dosage rather than of kind.

• Name the similarities the argument actually cites, then ask of each: is there a known connection between this feature and the property being projected, or is it merely conspicuous?
• List the differences between the two cases. Does any of them fall exactly where the projected property is determined?
• Construct a third case that matches every cited similarity but plainly lacks the projected property. If one is easy to build, the analogy is not carrying the conclusion.
• How much weight is the conclusion being asked to bear — a hunch worth checking, or a decision? Would the analogy still be offered if the stakes were reversed?

Legitimate when: Analogical reasoning is legitimate and indispensable — it is how case law, medicine, engineering, and animal research all proceed. It is strong exactly when the shared features are known to bear on the projected property, when the relevant differences have been surveyed and found not to touch it, and when the conclusion is stated no more strongly than that support permits. A modest conclusion drawn from a structural or causal similarity is a good analogical argument; the identical comparison pushed to a confident, high-stakes conclusion on the strength of surface likeness is a weak one.

Ancient: Aristotle does not name a fallacy of weak analogy; the ancient discussion of the weakness sits in his account of the objection. Arguments that reason from a parallel case rest on what holds for the most part rather than of necessity, and one standard way of meeting them is the objection drawn from a similar case — as against the claim that the injured always hate, one replies that the benefited do not always love. The strategy is exactly the modern one: produce a case matching the cited parallel in which the projected property fails. Confidence is medium because this is the ancient treatment of the underlying defeasibility, not an attested ancient naming of the error. — Aristotle, Rhetoric II.25 (on objection, enstasis)

Built from: Stanford Encyclopedia of Philosophy — Analogy and Analogical Reasoning · The Fallacy Files — Weak Analogy · Logically Fallacious — Weak Analogy · Internet Encyclopedia of Philosophy — Fallacies

Willed Ignorance

Refusing to look at something that might change your mind, and then carrying on as though your mind had been made up on the merits.

In the driveway, Nadia hands her father-in-law the mechanic's inspection sheet on the truck he wants to buy from a friend at church. He folds it in half without looking and puts it in his shirt pocket. "I don't need to read that. I've known Ray thirty years." It is not that he has read the sheet and thinks the mechanic is wrong. It is that the sheet stays folded, and by dinner he is telling everyone the truck is sound.

The tell: It is the physical refusal you notice before anything else — the paper pushed back across the table, the headphones coming off, the link left unclicked, always with a little smile. What follows is the strange part: a moment later they will speak about the thing they declined to see as if they had seen it and it was nothing. You feel like you're arguing with someone who has left the room but keeps talking.

The move: An arguer holds claim C. Evidence E bearing directly on C is available at low cost and is offered. The arguer declines to examine it — not by rebutting it, not by impeaching its source, but by refusing inspection, and characteristically on the ground that inspecting it might unsettle C. C is then maintained at its original confidence and presented to others as though the available evidence had been weighed and found wanting. The refusal thereby does the work of a premise: that C stands unless overturned by something the arguer has committed in advance to never looking at.

Why it fails: Confidence in a claim is only ever as good as the inquiry behind it, and the inquiry here was deliberately truncated at the point where it might have bitten. The arguer is not entitled to treat unexamined evidence as absent evidence, because the reason it went unexamined was precisely the risk that it told against them — which makes the refusal itself evidentially significant in the wrong direction. Two distinct defects are stacked. The first is a suppressed premise: the argument's real support is 'nothing I have consented to see contradicts C', which is far weaker than the 'nothing contradicts C' it is passed off as. The second is that the selection of what to look at is being made by the desired answer, so the arguer's evidence base is not a sample of the available evidence but a filtered one, and no amount of consistency inside a filtered base tells you anything about the world outside it. Note also what this is not: it is not a claim about the arguer's character. A person who refuses to look may still be right, and often is; what fails is the entitlement to the confidence, not the claim.

• Was specific, available, low-cost evidence offered and declined without being examined? Name the evidence and the moment it was refused.
• What ground was given for the refusal — that the source is unreliable (a reason), or that examining it might change their mind (not a reason)?
• Is the confidence being asserted higher than what the evidence they have actually looked at could support?
• Does the arguer treat the refused material as though it had been reviewed and dismissed — 'there's nothing in that' from someone who has not opened it?

Legitimate when: Declining to examine something is often correct and sometimes obligatory. Attention is finite, so triage is rational; sources with a track record of fabrication can be set aside on that ground; matters already settled by better evidence need not be relitigated on every fresh anecdote; a juror is instructed to ignore inadmissible material; and a person in grief may reasonably postpone reading a report. The move becomes this error only when the refusal is grounded in the fear that the material would work, and when the resulting belief is nonetheless carried at a confidence that only the refused inquiry could have earned.

Ancient: Aristotle holds that not all ignorance excuses. "Indeed, we punish a man for his very ignorance, if he is thought responsible for the ignorance," he writes, as penalties are doubled for the drunk, since the man had the power not to get drunk and his drunkenness caused his ignorance. He adds that we punish those ignorant of things in the laws they ought to know and that are not hard to learn, "and so too in the case of anything else that they are thought to be ignorant of through carelessness; we assume that it is in their power not to be ignorant, since they have the power of taking care." The passage is about culpability rather than about argument, but it is the ancient articulation of the idea this record rests on: ignorance you chose does not carry the protections of ignorance you suffered. — Aristotle, Nicomachean Ethics III.5

Built from: Logically Fallacious — Willed Ignorance · Wikipedia — Willful ignorance · Aristotle, Nicomachean Ethics III (MIT Internet Classics Archive) · Stanford Encyclopedia of Philosophy — Fallacies

Wishful Thinking

Believing something because you badly want it to be true, and then offering the wanting as though it were a reason.

Kitchen table, ten at night, a spreadsheet open between them. Dev has run the numbers on the house three times and every version leaves them ninety dollars short each month. "It'll be fine," Amara says. "My hours always pick up in spring. Something always comes through for us — it has every single time." Dev asks what happens if it doesn't, and she says she isn't going to think that way because thinking that way is how people talk themselves out of things. The spreadsheet has not changed. The offer goes in on Thursday.

The tell: The reason they give is about how much is riding on it rather than about whether it's so. There's a warmth in the room that makes disagreeing feel like betrayal, and you notice you've started arguing about whether you're being negative instead of about the numbers. If you find yourself defending your loyalty rather than your figures, the swap has already happened.

The move: A claim P is desirable to the arguer, or its denial is painful to them. The arguer's confidence in P is set by that desirability rather than by the evidence, and the resulting belief is then presented — to an audience or to themselves — as though it had been reached from the evidence. In its explicit form the desirability is stated outright as the reason: 'it would be unbearable if not-P, therefore P.' In its far commoner implicit form the desire operates upstream of the argument, deciding which evidence gets collected, which items are pressed hard and which are waved through, and where the inquiry is allowed to stop.

Why it fails: How good an outcome would be and whether it will occur are two independent matters, so the desirability of P supplies nothing whatever about P's truth; a premise about what the arguer wants sits in the place where a premise about the world is required, which is the plain relevance failure the emotional appeals all share. The implicit form is worse than the explicit one, because it leaves no visible bad step to point at: every individual move can look reasonable while the standard of evidence has been silently set lower for the welcome answer and higher for the unwelcome one, so the argument's asymmetry lives in the thresholds rather than in any stated premise. Two further consequences make it worth its own record. First, the arguer's own sincerity is no protection — this is a mistake people make most reliably about things they care about most, so the strength of feeling that P is not a check on it but a symptom. Second, because the belief was never fastened to evidence, disconfirming evidence arriving later tends not to loosen it, and the wishful believer characteristically ends up needing an additional story about why the world failed to cooperate.

• Would this same body of evidence persuade the arguer if the outcome were reversed — if what they hope for and what they fear traded places?
• Is the desirability of the outcome stated anywhere as though it were a reason? Look for 'it has to be', 'I refuse to believe', 'that just can't happen to us'.
• Was the unwelcome evidence held to a higher standard than the welcome evidence — more sources demanded, more caveats attached, more benefit of the doubt withheld?
• Ask the arguer to state the case against, in full and fairly. Can they do it, or does the attempt collapse into why it would be terrible if it were so?

Legitimate when: Hope, optimism, and acting on a favourable outcome under uncertainty are all legitimate and are not this move. Deciding to proceed with a venture whose odds you have honestly appraised as poor is a decision, not a belief, and is defensible on its own terms; so is a well-founded optimism where the evidence genuinely supports the good outcome. There is even a narrow class of self-fulfilling cases — confidence that measurably improves performance, trust that makes cooperation possible — where believing P raises the chance of P, and taking that effect into account is sound rather than wishful. The error is specifically the substitution: desirability standing in for evidence, and the resulting confidence being reported as a finding.

Ancient: Urging the Athenians to face what funding a war would actually cost them, Demosthenes dismisses the comfortable proposals for finding money that would require nobody to give anything up: "I think that in such proposals the wish is father to the thought, and that is why nothing is easier than self-deceit. For what each man wishes, that he also believes to be true." The remark is made about a live political decision rather than as a piece of logical theory, and it is the ancient world's cleanest statement of the move — belief following desire, and doing so most easily where the stakes are highest. — Demosthenes, Third Olynthiac 19

Built from: Wikipedia — Wishful thinking · Logically Fallacious — Wishful Thinking · Demosthenes, Third Olynthiac (translation) · Stanford Encyclopedia of Philosophy — Fallacies

Amphiboly

also known as: amphibolia, amphibology, syntactic ambiguity

A sentence that can be put together two different ways, and someone runs with whichever way happens to help them.

The new policy taped up by the walk-in at the restaurant reads: "Servers must not eat food prepared for guests in the dining room." Tomas, three weeks in, has been standing at the pass finishing a returned plate. When the manager writes him up he points at the sheet — it says in the dining room, and he wasn't in the dining room. The manager meant food prepared for the guests who are in the dining room. Both readings are sitting there in the same nine words, and the write-up now turns on which one the sign was.

The tell: You are both quoting the same sentence, word for word, and neither of you is cheating. That is the strange part — usually when people disagree about what was said, somebody is trimming. Here nobody is, and you keep saying the line back to each other slower and louder as if that would fix it.

The move: A sentence S admits two or more grammatical parses, P₁ and P₂, which yield materially different claims — through a modifier that can attach to either of two phrases, a pronoun with two possible antecedents, an unmarked scope for an adverb or quantifier, or a coordination that can be read as one item or two. Whatever supports S — the evidence, the authority, the speaker's actual intention — supports it only under P₁. The arguer draws a conclusion licensed only by P₂, treating the fact that the sentence was uttered as though it had settled the reading. Every word in S keeps a single dictionary meaning throughout; the ambiguity is structural, which is what separates amphiboly from equivocation.

Why it fails: A sentence with two parses is not one claim with a blurry edge; it is two claims wearing the same clothes, and only one of them was ever asserted. What licenses a conclusion is the claim, not the string of words, so the argument's second step is drawing on content that the first step never delivered — the arguer has silently chosen the reading that suits them and charged it to a speaker who chose the other one. The reason the move is persuasive is that nothing has been misquoted: the sentence really does say that, under a parse the grammar genuinely permits, so the ordinary defence of quoting accurately provides no protection at all. The test that exposes it is mechanical rather than interpretive. Write out each admissible parse as its own unambiguous sentence, then ask two questions: which of them does the evidence or the speaker actually support, and does the conclusion follow under every parse or only one? If the conclusion survives only under a parse the evidence does not support, the argument was carried by the syntax rather than by anything anyone said.

• Can the sentence be parsed two ways that differ in what they claim — not merely in tone? Write both out as separate, unambiguous sentences before answering.
• Which parse does the evidence, the authority, or the speaker's known situation actually support?
• Does the conclusion follow under every admissible parse, or only under one of them?
• Are all the individual words holding a single meaning throughout? If a word has shifted sense, the charge is equivocation instead; if only the grammar has shifted, it is amphiboly.

Legitimate when: Ambiguous sentences are everywhere and reasoning from them is unavoidable — contracts, statutes, scripture, and text messages all get read for a living. The move is legitimate whenever the parse is fixed rather than assumed: by context that only one reading fits, by the speaker's other statements, by an interpretive canon the parties accept, or by openly arguing for the reading before relying on it. What makes it an error is helping yourself to the favourable parse in silence, as though the sentence had only ever had one shape.

Ancient: Aristotle lists amphiboly second among the six ways of producing a false appearance of argument that depend on language: "ambiguity, amphiboly, combination, division of words, accent, form of expression." His illustrations are Greek sentences whose grammar admits two construals — most famously the line rendered in English as "I wish that you the enemy may capture," where the case endings leave open which party captures which, and also "There must be sight of what one sees: one sees the pillar: ergo the pillar has sight," where the genitive can name either the seer or the thing seen. The defect he is isolating is in the construction, not in the meaning of any single word. — Aristotle, On Sophistical Refutations, ch. 4

Built from: Aristotle, On Sophistical Refutations (MIT Internet Classics Archive) · The Fallacy Files — Amphiboly · Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Syntactic ambiguity · Internet Encyclopedia of Philosophy — Fallacies

Fallacy of Combination

also known as: compositio, combination of words

Someone agrees to two things one at a time, and you hold them to having agreed to both at once.

At breakfast Ade's mother says two things, several minutes apart. He can take the car on Friday. And yes, he can go to the show in the city. Friday night the car is gone and so is he. "You said both," he tells her on the phone from a parking garage in the city. She had meant the car for the twenty minutes to work and back, and the show by train with everybody else. She agreed to each of those and never once to the pair, and the sentence he built out of the two of them is one she would have turned down flat on the spot.

The tell: Every piece of what they're quoting is something you genuinely said, which is why you can't find the lie. The wrongness is in the seam. If you catch yourself thinking "yes, but not together" and then hesitating because it sounds like you're weaseling, that hesitation is the move working on you.

The move: A sentence containing two or more expressions admits a divided reading, on which each is asserted separately, and a composite reading, on which they fall jointly under a single operator — most often the scope of a modal ('can', 'must', 'is able to', 'is permitted'), an adverb, or a quantifier. The premise is granted under the divided reading, where it is true. The arguer then draws a conclusion licensed only by the composite reading, so that abilities, permissions, or predicates conceded one at a time are treated as though they had been conceded together. This is Aristotle's sunthesis, third in his list of the fallacies that depend on language.

Why it fails: The composite reading is not a fuller statement of the divided one; it is a different and ordinarily much stronger claim, because the operator has changed what it governs — from each conjunct in turn to their conjunction. 'He can walk' and 'he is sitting' are jointly satisfiable facts about a man; 'he can walk-while-sitting' asserts a single compound ability that no fact about him supports. Nothing in the premise fixes the composite reading, so the argument's whole force comes from the sentence's willingness to be read either way, which is a fact about how words combine in a string rather than about the world the words are describing. The error is therefore one of scope rather than of quantity: no step adds an unsupported item, and every word retains its meaning; what silently widens is how far the modal reaches. It is exposed by bracketing — mark explicitly what each 'can' or 'must' governs and rewrite each reading as its own sentence — after which the premise is seen to have asserted the weaker one all along and to have no bearing on the stronger.

• Rewrite the premise twice: once with the items taken separately, once with them taken jointly under the same 'can', 'must', or adverb. Which one was actually granted?
• Does the conclusion need the joint reading? If the joint reading is deleted and only the separate claims remain, does anything survive?
• Bracket the scope of every modal and adverb explicitly. Does the scope in the conclusion match the scope in the premise, or has it widened?
• Is there any independent reason to think the two things can be done, or hold, at the same time — as opposed to each being possible on its own?

Legitimate when: Many predicates and abilities do combine, and combining them is ordinary sound reasoning: from 'he can whistle' and 'he can walk' it is normally fine to conclude he can walk and whistle, because there is no interference between the two. What is required is a reason to think the items are jointly satisfiable — no shared resource, no mutual exclusion, no contradiction in the compound. The step is legitimate whenever that joint satisfiability is established or obvious, and an error whenever it is merely inherited from a sentence that could be read either way.

Ancient: Aristotle names combination third among the six fallacies that depend on language — "ambiguity, amphiboly, combination, division of words, accent, form of expression" — and illustrates it with sentences whose truth flips according to whether the words are taken together or apart: "A man can walk while sitting, and can write while not writing." Taken divided, the claim is true and unremarkable: a seated man retains the ability to walk. Taken combined, it asserts an ability to walk-while-seated, which is false. Its partner, the fallacy of division, runs the other way, and Aristotle's example there is that "5 is 2 and 3," and odd. — Aristotle, On Sophistical Refutations, ch. 4

Built from: Aristotle, On Sophistical Refutations (MIT Internet Classics Archive) · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies · The Fallacy Files — Composition

Figure of Speech Fallacy

also known as: figura dictionis, form of expression

Two words are shaped alike, so it gets assumed that the things they name work alike.

Sunday dinner, and Uncle Ray has views about Cass's anxiety. "Worrying is something you do," he says, pointing his fork at her. "Same as jogging. Same as smoking. Nobody's a victim of jogging — you do it, so you can quit doing it. Just stop." It sounds airtight, and part of why it sounds airtight is that 'worrying' and 'jogging' come out of the mouth in exactly the same shape. But one of them is a thing Cass decides to start on a Tuesday morning and the other is what her body does at three in the morning whether or not she was consulted.

The tell: The point lands like a small trick with words, and it lands a beat before you've understood it. If you replay it, you notice that the whole thing turned on two words rhyming in structure, and that nobody said anything about the actual things. The room went quiet because the sentence was well built, not because anyone learned something.

The move: Two expressions share a grammatical or figurative form — the same ending, the same part-of-speech shape, the same construction, the same metaphor — and the argument treats that shared form as evidence that what they name belongs to the same logical category, so that what holds of one is transferred to the other. Aristotle's own cases (schēma tēs lexeōs, rendered in Latin as figura dictionis) turn on Greek morphology: a thing of one gender expressed with the termination of another, or a word naming a quality wearing the form of a word naming an action, whereupon the quality is reasoned about as though it were something done. The modern extension keeps the shape but moves it up a level: a metaphor's grammar is taken to license the moves the literal category would license.

Why it fails: Grammatical form tracks the history and convenience of a language, not the joints of the world, so the fact that two expressions inflect alike or sit in the same slot of a sentence is a fact about words and carries no information about what the words name. When the only bridge between two steps of an argument is that the two expressions look alike, the bridge is orthographic, and the conclusion is being supported by a coincidence of endings. The failure is a category error with a persuasive disguise: because the sentence is well formed, it sounds as though it must be well reasoned, and the audience's fluency in the grammar is mistaken for their grasp of the claim. It is exposed by paraphrase. Restate each step with the shared form or figure removed — say plainly what happens rather than using the word that shares the shape — and if the connection between the steps evaporates, the form was doing all the work. Aristotle's illustrations depend on features of Greek that do not survive translation, so modern instances of this record are extensions of his pattern rather than reproductions of his examples; the pattern extends cleanly, but that limitation should be stated rather than hidden.

• Do the two expressions share only a grammatical shape or a figure of speech, or do they also share a category — is one an action and the other a state, one a thing and the other a quality?
• Paraphrase away the shared form or metaphor and restate the argument in plain description. Does the connection between the steps survive?
• Is a predicate that belongs to concrete things (having, losing, getting, removing, treating) being applied to an abstraction purely because its noun behaves grammatically like a concrete noun?
• Would a speaker of a language that expresses these two ideas with unrelated words find this argument compelling at all?

Legitimate when: Grammatical form is sometimes a genuine guide, because languages do encode real distinctions, and an argument may lean on it where the encoding is reliable and the reliance is stated. Metaphors and models are likewise indispensable and legitimately transfer conclusions — an economist may reason about a market as a system, an epidemiologist about a rumour as a contagion — provided the correspondence between the two domains is spelled out and defended item by item, so that the transfer rests on the argued mapping rather than on the fact that one can say the sentence.

Ancient: Form of expression is the sixth and last of the fallacies Aristotle says depend on language. Some arise, he writes, "owing to the form of expression used, when what is really different is expressed in the same form, e.g. a masculine thing by a feminine termination, or a feminine thing by a masculine, or a neuter by either a masculine or a feminine." His standing illustration is that "'flourishing' is a word which in the form of its expression is like 'cutting' or 'building': yet the one denotes a certain quality... while the other denotes a certain action." The mistake is to let the shared shape of the word settle what kind of thing it names — for Aristotle the leading cause of metaphysical error. — Aristotle, On Sophistical Refutations, ch. 4

Built from: Aristotle, On Sophistical Refutations (MIT Internet Classics Archive) · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies · Aristotle's account of fallacies in the Sophistical Refutations (The Non Sequitur)

Secundum Quid

also known as: secundum quid et simpliciter, qualified and unqualified statement, converse accident

Taking something that is only true in one particular respect and repeating it flat out, with the 'in this respect' quietly gone.

Six days after the knee replacement, Grandpa Whit announces at the table that the surgeon cleared him to walk. This is true. What the surgeon said was that he could walk on level ground, indoors, ten minutes at a stretch, with the frame, and only if there was no swelling that morning. By Saturday he is down at the lake with a fishing pole and his grandson, on a gravel slope, because the sentence he carried out of that appointment was "cleared to walk," and every condition on it fell off somewhere between the exam room and the parking lot.

The tell: You recognise your own words coming back at you and they've gotten bigger. There's a specific small lurch to it — you want to say "that's not quite what I said" while knowing that every word of it is what you said. What's missing is the part after the comma, and the part after the comma was the whole point.

The move: A claim holds only under a qualification — in a certain respect, at a certain time, in a certain manner, under certain circumstances, or for a certain part. In the primary direction (a dicto secundum quid ad dictum simpliciter) the qualification is dropped and the claim is asserted flat out: 'S is P in respect R' becomes 'S is P.' In the converse direction (a dicto simpliciter ad dictum secundum quid) an unqualified rule is applied to a case that falls under one of its recognised exceptions. Aristotle's own treatment in Sophistical Refutations ch. 5 is chiefly of the first: he warns that 'to be x' and 'to be' absolutely are not equivalent, and that a disputant may extract an admission carrying its qualification and then trade it in for the unqualified one.

Why it fails: A qualification is not a hedge attached to a claim; it is part of the claim, and stripping it produces a different and stronger statement that the evidence for the original never touched. Because the qualified version really was granted, the arguer can quote the concession accurately while reasoning from something the respondent never conceded, which is what makes the move so hard to answer in the moment — the words are right and the content has changed underneath them. The characteristic product is an apparent contradiction manufactured out of two compatible truths: a man white in respect of his teeth and black overall is both white and not white only if both qualifications are discarded, and the discarding is the arguer's act, not the world's. In the converse direction the damage runs the other way: a rule that holds for the standard case is enforced against precisely the situation its exceptions were written for, so the case is decided by the rule's slogan rather than by the reasons the rule exists. Both directions are failures of presumption rather than of evidence — the missing qualification is smuggled out of, or into, the premises without ever being argued.

• State the claim in its original, fully qualified form. What respect, time, manner, quantity, or circumstance was attached to it?
• Does the step being made still hold with the qualification restored — or does it need the flat, unqualified version?
• If a general rule is being applied, does the rule admit recognised exceptions, and does this case fall inside one of them?
• Is an apparent contradiction being generated? If so, check whether the two sides were each true under a different qualification that has been discarded.

Legitimate when: Dropping a qualification is fine when it is genuinely idle — when the claim holds in every relevant respect, so the qualifier adds nothing — and generalising from qualified cases is fine when the qualification has been shown not to matter for the property in question. Equally, applying a general rule to a particular case is exactly what rules are for; the step is sound whenever the case has been checked against the rule's stated exceptions and found not to fall under them. What is required is that the qualification be examined and released, rather than lost in transit.

Ancient: Among the fallacies not depending on language Aristotle sets those that turn on an expression being used absolutely or in a certain respect, warning that "'to be x' and 'to be' absolutely" are not the same, and likewise for their denials. His illustration: "Suppose an Indian to be black all over, but white in respect of his teeth; then he is both white and not white." The sophistical procedure he has in view is procedural as well as logical — secure the qualified admission first (that the Ethiopian is white in respect of his teeth), then argue from it as though the unqualified statement had been granted. — Aristotle, On Sophistical Refutations, ch. 5

Built from: Aristotle, On Sophistical Refutations (MIT Internet Classics Archive) · Wikipedia — Secundum quid · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies

Ignoratio Elenchi

also known as: ignoratio elenchi, ignorance of refutation, irrelevant conclusion

Proving something true, and then behaving as though it had settled a completely different question.

At the parent-teacher conference Renata asks one thing: is Mateo going to pass biology. The teacher talks for eleven minutes about how much he has grown this year — how he took the new kid under his wing in September, how careful he is at the lab bench, how much she personally looks forward to that period. All of it is true and Renata is glad to hear every word. She walks back out to the parking lot still not knowing whether her son is going to pass biology.

The tell: You come away warmed and still hungry. Nothing you were told was false, nobody was rude to you, and you cannot point to the moment it went wrong — but the question you brought in is still sitting there untouched. If you have to ask yourself "wait, did they answer me?", the answer is usually no.

The move: A dispute fixes a thesis T. To refute T is to establish its contradictory — of the same attribute, of the same subject, and, in Aristotle's list of conditions, in the same respect, relation, manner, and time in which T was asserted. The arguer offers an argument, frequently an impeccable one, whose conclusion is some Q that fails at least one of those conditions: Q concerns a neighbouring subject, or the same subject in a different respect, or the same claim at a different time, or a claim weaker or stronger than the contradictory of T. Q is then presented as though the dispute had been settled by it.

Why it fails: The defect is not in the argument but in its address. Every step from premises to Q may hold, and typically does, which is why this move survives scrutiny that would kill a weaker one: examine the reasoning and it checks out, and the checking distracts from the fact that the reasoning was never aimed at T. Support is a relation between an argument and a specific claim, so an argument that establishes Q supports T's denial only if Q entails it, and where the two differ in respect, time, manner, or subject, that entailment is absent and nothing at all has been done to T. What makes the conditions in Aristotle's definition load-bearing is that each of them can be violated while every word of the original thesis is preserved — the same sentence, argued about at a different time or in a different respect, is a different claim. Aristotle also gives this record an unusual status: in Sophistical Refutations ch. 6 he argues that the whole apparatus can be referred back to it, since "all the types of fallacy fall under ignorance of what a refutation is." That makes ignoratio elenchi the genus rather than one species among many, and it is why a charge of it should be made only when no narrower record fits.

• Write down the thesis actually in dispute, in one sentence, before reading the reply. Now write down what the reply establishes. Are they the same claim negated, or two different claims?
• Grant the reply's conclusion outright and in full. Is the original question still open? If it is, the reply was aimed elsewhere.
• Check the four conditions one at a time: same subject, same respect, same manner, same time. Which one does the reply fail?
• Is a narrower record available — a strawman restating of the thesis, a red herring introduced to divert, an attack on the arguer? If one fits, charge that instead; this record is the genus.

Legitimate when: Proving something adjacent to the thesis is legitimate and constant in real argument: establishing a lemma on the way, undermining a premise the thesis depends on, or showing that the dispute rests on a false presupposition and so does not arise. Q also settles the matter outright whenever Q genuinely entails the contradictory of T. The requirement is only that the arguer say what they are doing — offer Q as a step, or as a reframing, or as an entailment with the entailment shown — rather than presenting Q as though it were the refutation itself.

Ancient: Aristotle defines the standard the move fails: "To refute is to contradict one and the same attribute — not merely the name, but the reality — and a name that is not merely synonymous but the same name — and to confute it from the propositions granted, necessarily, without including in the reckoning the original point to be proved, in the same respect and relation and manner and time in which it was asserted." Arguments that overlook any one of these conditions refute something other than the thesis at issue. In ch. 6 he goes further and makes this the master diagnosis: "All the types of fallacy, then, fall under ignorance of what a refutation is." — Aristotle, On Sophistical Refutations, ch. 5 (and ch. 6)

Built from: Aristotle, On Sophistical Refutations (MIT Internet Classics Archive) · Wikipedia — Irrelevant conclusion · Stanford Encyclopedia of Philosophy — Fallacies · Internet Encyclopedia of Philosophy — Fallacies

False Cause

also known as: non causa pro causa, false cause fallacy

Two things happening close together in time doesn't mean one caused the other — but this trick acts like it does.

In the group chat someone posts, 'Every time I wear my lucky socks, our trivia team wins. Socks are staying on forever now.' The team also happens to have its best player show up on sock nights. Nobody checks that — the socks get all the credit.

The tell: Two things happened near each other in time, and suddenly one gets blamed or praised for causing the other — with nobody asking whether something else was really going on, or whether it could be the other way around, or just plain luck.

The move: A is observed alongside, or shortly before, B (mere correlation, temporal succession, or co-occurrence); therefore A causes B — with the causal direction, the mechanism connecting them, and every rival explanation (reverse causation, a third factor producing both, coincidence, or regression to the mean) left unexamined.

Why it fails: Correlation and succession are consistent with several different causal stories — A causes B, B causes A, some C causes both A and B, or the link is coincidental / an artifact of small samples and natural fluctuation — and the argument never does the work of ruling the rivals out. Nothing has actually identified a mechanism by which A produces B, nor supplied the counterfactual (what would B have done if A had not occurred?) that a genuine causal claim requires. The observation is real; only the causal interpretation of it is unearned.

• Does the evidence show more than 'A happened, then B happened' or 'A and B tend to occur together' — is there an actual mechanism linking them?
• Could the arrow run the other way (B causing A) or could a third factor be driving both A and B independently?
• What would have happened to B without A — is there any comparison case, control group, or counterfactual, or just the one observed instance?
• Is this a coincidence or small-sample fluke that would vanish with more data, more time, or a comparable baseline (e.g., regression toward the mean)?

Legitimate when: The causal claim is backed by more than co-occurrence — a controlled or natural experiment, a documented mechanism, temporal precedence combined with ruled-out confounders, or a dose-response pattern — so that rival explanations (reverse direction, common cause, coincidence) have actually been checked and excluded rather than merely ignored.

Ancient: Aristotle names a refutation move he calls treating the non-cause as the cause (non causa pro causa): in a reductio-style argument, blaming an absurd result on some starting claim that actually played no part in producing it. This original sense is narrower than, and the historical ancestor of, the modern everyday habit of blaming a correlated or preceding event for something it may not have caused at all. — Aristotle, On Sophistical Refutations, ch. 5

Built from: Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Questionable cause · Wikipedia — List of fallacies

Fallacy of the Consequent

also known as: fallacia consequentis, consequent fallacy

Reading a one-way sign backwards — rain leaves the ground wet, so wet ground must mean it rained.

Priyanka gets home to a chewed sneaker in the hallway and walks the beagle straight into her crate. The whole thing takes about a second and does not feel like thinking at all: when Biscuit gets into something there is a wrecked shoe, and here is a wrecked shoe. Her eleven-year-old nephew had been over that afternoon with craft scissors and a project involving "leather." The dog spent the whole evening in the crate for something she watched somebody else do.

The tell: The answer arrives before the question does. There's no moment of weighing — you see the mess and you already know who, and the certainty has the texture of having seen it happen rather than of having worked it out. The give-away is that if somebody asks what else could have caused it, you can usually come up with something in about four seconds, which means the four seconds were available all along and you skipped them.

The move: Aristotle's fallacia consequentis is the supposition that a one-way relation of consequence converts. Given that whenever A holds B necessarily holds, the arguer takes it that whenever B holds A necessarily holds — and, running the same conversion in negative dress, that where A fails B fails too. In the ancient setting the relation is most often a sign-relation rather than a stated conditional: something is regularly attended by something else, and the attendant is then read backwards as proof of the thing. Aristotle classifies it as a species of the fallacy of accident, on the ground that following upon a thing is an attribute of that thing rather than the thing itself, and he notes its heavy use in rhetorical argument from signs.

Why it fails: A relation of consequence is directional, and the direction is the whole content of the claim: 'A brings B' says nothing about what else may bring B. Wherever B can arise from more than one source — and for signs it almost always can — the presence of B is compatible with A's absence, so the premises can hold with the conclusion false, and the argument establishes nothing. The move is seductive because a sign that reliably follows its cause feels like an observation of the cause rather than of an effect, and in ordinary perception it functions that way, which is why Aristotle chose examples from medicine and weather where the backwards reading is almost automatic. Two consequences of his framing matter for how the record is used. First, his version is wider than the modern schema that descends from it: it covers unstated sign-inferences and everyday expectations, not only arguments containing an explicit if-then, and it includes the negative direction as the same error seen from the other side. Second, because the strength of a sign-inference depends on how often the sign occurs without the thing, the ancient version shades into a question of relative frequency that the bare form cannot settle — the reason the same shape can be a serious defect in one case and a decent working guess in another.

• Does the argument run the stated relation backwards — from the thing that follows to the thing that would have brought it?
• Name at least one other way the observed result could have come about. If naming one is easy, the backwards reading is carrying no weight.
• Is the relation being relied on a sign-relation ('where there's X there's usually Y') rather than a stated rule? If so, ask how often the sign appears without the thing.
• Check the negative twin as well: is the argument concluding that because the first thing is absent the second must be too? That is the same conversion in the other direction.

Legitimate when: Reading a sign backwards is legitimate and constant when the sign is rare among the alternatives — when the observed result would be surprising unless the suspected source were present, so its presence really does raise the odds. That is diagnosis, tracking, and detective work, and it is respectable exactly to the degree that the competing explanations have been surveyed and found unlikely. The relation also converts outright when it genuinely runs both ways and the argument says so: with a biconditional, or with a definition, or with an exhaustive list of sources of B, the step from B to A is sound.

Ancient: Aristotle isolates the error directly: "For whenever, suppose A is, B necessarily is, they then suppose also that if B is, A necessarily is." His weather case is the ancestor of every modern textbook version — "since after rain the ground is wet in consequence, we suppose that if the ground is wet, it has been raining; whereas that does not necessarily follow" — and he gives a medical one alongside it, that a feverish man is hot but a hot man need not be feverish, and a case from perception, that "people often suppose bile to be honey because honey is attended by a yellow colour." He treats the whole class as a species of accident and notes that it is the standard machinery of arguments from signs. — Aristotle, On Sophistical Refutations, ch. 5

Built from: Aristotle, On Sophistical Refutations (MIT Internet Classics Archive) · Stanford Encyclopedia of Philosophy — Fallacies · Wikipedia — Affirming the consequent · Internet Encyclopedia of Philosophy — Fallacies