Why it matters

Some arguments are false. Other arguments are broken — and the two are not the same thing. An argument can reach a true conclusion through reasoning that does not actually support it, the way a stopped clock is right twice a day; and an argument can be impeccably constructed yet rest on a premise that happens to be wrong. A coherence audit ignores the second question entirely and goes after the first. It does not ask “is this conclusion true?” It asks “does this argument hang together — do the parts connect, does the conclusion actually follow, or is something quietly missing, doubled back on itself, or sliding around under your feet?” Most arguments that feel wrong are wrong here, at the joints, and most people can feel the wrongness long before they can name it. This mode names it.

For example: someone tells you “cryptocurrency can’t be a Ponzi scheme, because a Ponzi scheme by definition requires a fraudulent promoter, and Bitcoin has no promoter — so the regulators are categorically confused.” It sounds airtight. But walk the joints. The argument leans on a definition of “Ponzi scheme” narrow enough to require a single named operator, then treats that stipulated definition as if it settled a question about real-world harm — that is a key term doing two different jobs in one breath. And the conclusion (“regulators are confused”) does not follow from the premise even if the premise held: showing a thing is not one specific kind of fraud says nothing about whether it is some other kind. The conclusion might still be true — maybe the regulators really are confused — but this argument has not earned it. That gap between “the conclusion is false” and “this argument does not establish its conclusion” is the whole of what a coherence audit protects.

  • What it reveals. Whether an argument is internally sound in its wiring — free of self-contradiction, free of gaps between premises and conclusion, free of premises that secretly assume the conclusion, and free of key terms that shift meaning mid-argument — independent of whether its conclusion happens to be true.
  • How it changes the read. You stop asking “do I agree with this?” and start asking “does this actually follow?” — separating your reaction to the conclusion from the integrity of the path that was supposed to get you there.
  • When to foreground it. You have a specific argument in front of you — a paragraph, a claim-with-reasons, an op-ed’s central move — and you want a neutral structural check: does it hold together, where exactly does it break, without taking a side on the conclusion.
  • What you’d miss without it. That a conclusion you agree with can be propped up by reasoning that collapses on inspection — and that endorsing the argument because you like the conclusion is exactly how a bad argument survives to be used again.
  • Where it misleads. Pushed past its remit it slides into the fallacy fallacy — treating a broken argument as proof the conclusion is false, when a flawed path to a claim leaves the claim’s truth entirely open and demanding to be settled some other way.

How it works

Start with the move that surprises people: a coherence audit is not a soundness check. In logic those are two different tests, and keeping them apart is the whole trick. An argument is sound when its premises are true and its logic is valid. A coherence audit deliberately sets the truth half aside and tests only the second half — does the thing hold together as a piece of reasoning, whatever the facts turn out to be? That is why an argument can pass this audit and still be false (true wiring, a wrong input feeding it), and why an argument can fail this audit and still reach something true (a right answer arrived at by a broken path). The audit checks the wiring, not the current running through it.

A useful picture: think of an argument as a small machine. Premises go in, a conclusion comes out, and inside are the connections that are supposed to carry you from one to the other. A coherence audit opens the casing and inspects four specific kinds of fault in the wiring — and only these, because these are the faults that live in the structure rather than in the facts.

The first fault is internal contradiction: does the argument assert X in one place and not-X in another? An argument that needs both “the data are unreliable” and, two sentences later, “the data prove my point” has shorted itself out — it cannot be repaired by checking the world, because it disagrees with itself. This is the oldest rule in the book, the law of non-contradiction: a statement and its denial cannot both hold at once, and any argument that quietly relies on both has failed before the facts are even consulted.

The second fault is a gap: does the conclusion actually follow, or is there a missing step? Most arguments leave steps unstated — that is normal, even efficient — but some leave out a step that is load-bearing, an assumption the whole inference silently rides on. “Sellers are accepting price cuts, so high rates are working, so the central bank should hold” hides a step: that the price cuts were caused by the rates and not by something else, and that “working” in the past tense licenses “hold” in the future tense. Name the missing step out loud and the argument either earns it or visibly cannot. The audit’s job is to drag those silent steps into the light so you can see whether they hold.

The third fault is circular reasoning: does a premise smuggle in the very conclusion it was supposed to prove? “We can trust this source because it’s reliable, and we know it’s reliable because we trust it” travels in a perfect circle — it feels like an argument because it has the shape of one, premises and a conclusion, but the conclusion was already sitting inside the premise, so nothing was ever established. The tell is that you cannot accept the premise without already accepting the conclusion.

The fourth fault is equivocation: does a key term shift meaning partway through? This is the crypto example from earlier. A word like “Ponzi scheme” or “inflation” or “natural” gets used in a narrow technical sense to win one step and a broad everyday sense to win the next, and the argument coasts across the seam as if the word meant one thing throughout. Pin the term to a single definition and one of the steps stops working. The fault is invisible until you force the word to hold still.

Here is the discipline that makes the whole thing trustworthy, and it is the easiest one to lose. When the audit finds a fault — a contradiction, a gap, a circle, an equivocation — it has shown that this argument does not establish its conclusion. It has not shown that the conclusion is false. Treating “your argument is broken” as “you are wrong” is itself a fallacy (the fallacy fallacy, or argumentum ad logicam): a bad argument for a claim leaves the claim exactly where it was — possibly true, possibly false, simply not yet earned by this route. So the audit lands its verdict in a precise form — “this argument as given does not support its conclusion, because [the specific structural fault]; whether the conclusion is nonetheless true is a separate question, to be settled by other means” — and stops there. That restraint is not timidity. It is the difference between checking an argument and merely disagreeing with it.

Framework & implementation

Output contract

The deliverable is a fixed set of sections, so the audit is checkable rather than a free-form opinion. Charitable Reconstruction states the argument in its strongest fair form, with implicit premises made explicit. Per-Move Breakdown decomposes each inferential step into its Toulmin parts (and, where the move fits one, its Walton scheme with the critical questions marked satisfied / defeated / unaddressed), so the warrant — the silent connecting rule — is forced into view for every step. Contradictions lists any place the argument asserts something and its denial. Gaps names each missing load-bearing step where the conclusion does not follow without an unearned assumption. Circularity flags any premise that presupposes the conclusion. Equivocations identifies any key term that shifts meaning across the argument. Each named-fallacy finding carries full four-part substantiation — the quoted text, the inferential move it sits in, the principle the move violates, and the reason it fails here and not merely in the abstract — because a bare fallacy label without all four parts is treated as unsupported and dropped. A holds-or-fails verdict per move records whether each step survives. And the overall coherence verdict renders the load-bearing separation once at the document level: whether the argument as given establishes its conclusion, stated explicitly as a judgment about the argument and never about the truth of the claim.

Origin and evidence

The audit draws on two streams of twentieth-century logic. The oldest piece is the law of non-contradiction — that a proposition and its negation cannot both be true at once — which runs from Aristotle through the whole of classical logic and supplies the test for the contradiction fault. The notion that an argument can be assessed for internal fit apart from the truth of its inputs echoes coherentism, the family of views in epistemology on which a belief’s standing depends on how well it hangs together with the rest of a system rather than on a one-by-one correspondence with the facts; the audit borrows the spirit — test the connections — without taking on coherentism’s larger claims about truth. The working machinery comes from the mid-century revival of informal logic and fallacy theory: Stephen Toulmin’s model of argument (claim, grounds, warrant, and the rest) gave the field its parts-list for decomposing real-world inferences; C. L. Hamblin’s Fallacies (1970) reopened the serious study of how arguments go wrong by exposing the textbook fallacy lists as theoretically thin; and that revival is consolidated in working critical-thinking texts such as Trudy Govier’s A Practical Study of Argument (1985) and Douglas Walton’s Informal Logic (1989), which is also the source of the argumentation-scheme catalog and its critical questions. The four-fault frame — contradiction, gap, circularity, equivocation — is the practical distillation those traditions converge on.

Applications and common uses

  • Op-ed and commentary checking. Take the central argumentative move of a column and test whether it actually follows — separating a conclusion you might share from reasoning that may not support it.
  • Policy-claim scrutiny. Pressure-test the “because” in a policy argument (“we should do X because Y”) for hidden load-bearing steps and terms that shift meaning between the diagnosis and the prescription.
  • Spotting the term that slides. Catch equivocation in debates that turn on a single contested word — “inflation,” “natural,” “fraud,” “free speech” — where the argument quietly trades a narrow sense for a broad one.
  • Self-check before you publish. Run your own argument through the four faults before committing to it, so you find the gap or the circle yourself instead of having a critic find it for you.
  • Teaching and learning to argue. Make the invisible parts of an inference visible — especially the unstated warrant — as a way of learning what it actually takes for a conclusion to follow.

Failure modes and when not to use it

  • The fallacy fallacy. Sliding from “this argument is broken” to “this conclusion is false.” The guard is structural: the overall verdict is fixed in the form the argument does not establish its conclusion; its truth is a separate question — the mode is built so it cannot quietly cross that line.
  • Manufacturing gaps. Calling every unstated step a “gap.” Arguments compress, and most omitted steps are harmless; only a load-bearing missing assumption is a real gap. The cross-adversarial second analyst exists largely to throw out the false positives.
  • Fallacy-label spray. Tagging moves with named fallacies that do not really fit. The four-part substantiation requirement — quote, move, principle, why-it-fails-here — is the filter; a label that cannot earn all four parts is dropped rather than asserted.
  • Mistaking a fair definition for equivocation. Not every use of a technical term is a meaning-shift; a term is only equivocating if it does different work in different steps. The audit must show the seam, not just note that a word is technical.

When not to reach for it. Coherence Audit is the lightest tool in its territory, and three nearby jobs belong elsewhere. When you want the argument taken fully apart into its complete premise / warrant / backing skeleton and rebuilt — a thorough molecular reconstruction rather than a structural spot-check — that is argument-audit, the heavier mode this one sits beneath. When the real problem is not the wiring but the framing — the assumptions baked into how the question itself is posed, the words that pre-load the answer — that is frame-audit. And when you suspect the artifact is not merely a weak argument but is functioning as propaganda — engineered to move you rather than to reason with you — that is propaganda-audit. Coherence Audit checks whether an argument hangs together; it does not reconstruct it in full, interrogate its frame, or diagnose its persuasive intent.

  • Argument Audit — the heavier molecular sibling this mode sits beneath: when you want the full premise / warrant / backing reconstruction and an integrated analysis rather than a structural spot-check, that is the escalation route.
  • Frame Audit — the sideways route for when the trouble is not the argument’s wiring but the framing it is built on — the assumptions baked into how the question is posed.
  • Propaganda Audit — the adversarial-stance sibling for when the artifact may be engineered to persuade rather than to reason, and the question is its function, not just its coherence.
  • Toulmin Decomposition and Walton Schemes and Critical Questions — the two lenses this mode hosts: the parts-list that surfaces the unstated warrant in every move, and the catalog of real argument forms whose critical questions locate each form’s known weak points.