Why it matters

Arguments are not snowflakes. The vast majority of the reasoning you meet falls into a few dozen recognizable types — appeal to an expert, argument from a sign, reasoning from cause to effect, the slippery slope — and each type comes with its own short checklist of questions that, answered the wrong way, knock it down. Learn the type and you inherit the checklist; you stop arguing on instinct and start probing exactly where this kind of argument tends to break.

For example: an ad says “four out of five dentists recommend this gum.” It feels authoritative, and you sense something is off, but flailing for a rebuttal you reach for “well, that’s just an ad.” Name the type instead — this is an appeal to expert opinion riding on an appeal to popular opinion — and the checklist writes your rebuttal for you: Which dentists? Recommend it over what? Were they paid? Is “most dentists recommend it” even evidence the gum is good, or just that it clears a low bar? The argument hasn’t been called names; it has been tested at its actual joints.

  • What it reveals. Which named argument form a piece of reasoning instantiates, and the specific critical questions whose negative answers would defeat it — the joints where this type of argument characteristically fails.
  • How it changes the read. You stop asking “do I buy this?” and start asking “what kind of argument is this, and has it answered the questions that kind of argument has to answer?”
  • When to foreground it. Any argument that fits a familiar shape — citing an authority, drawing an analogy, predicting a chain of consequences, inferring intent from a sign — especially when it feels persuasive and you want to locate why it does or doesn’t hold.
  • What you’d miss without it. That a presumptive argument is defeasible by design — it was only ever meant to hold “unless,” so an unanswered critical question makes it under-determined, not refuted; treating it as proven (or as debunked) both misread what it claimed.
  • Where it misleads. Misclassify the form and you import the wrong checklist and produce a confident but off-target audit; and a checklist can be turned on a strawman — auditing a reconstruction the arguer never actually made.

How to invoke it in Ora

You have an argument and you want to know whether the reasoning actually holds up — not whether you agree with the conclusion, but whether the premises genuinely support it.

Paste the argument and ask:

“Coherence audit: does this argument actually hold together? Check the reasoning, not the conclusion.”

This rides inside the Coherence Audit analysis. Ora first rebuilds the argument at its strongest and most charitable, then breaks each step into its parts (the claim, the evidence, the unstated rule connecting them) and — where a step is one of the recognized argument types — names the type and runs its critical-question checklist, marking each question supported, defeated, or merely unaddressed. The Walton-schemes lens is one of the two reasoning tools the audit is built on (alongside the Toulmin model): it supplies the named types and their checklists, so the audit can test each step at the exact joints that kind of argument is known to fail.

One thing to know: phrases like coherence audit, does this hold up, check the reasoning, fallacy check, or Walton / critical questions are what route you here. Naming the lens alone — “use Walton’s schemes” — does not route; describe the argument and ask whether the reasoning holds.

Give it the actual argument, premises and all, because the lens works by matching the reasoning to a named scheme; a conclusion with no visible premises gives it nothing to classify.

One thing Ora won’t do: confuse a shaky argument with a false conclusion. It keeps “this step is defeated” separate from “this question is merely unaddressed,” and it says plainly that an argument can fail while its conclusion stays perfectly true — the conclusion is simply not established by this argument.

How it works

Here is a small puzzle that has bothered logicians for a long time. Take the argument: “Dr. Chen is a climate scientist, and she says the data show accelerating warming, so it’s probably accelerating.” Is that valid? By the rules of strict logic — the airtight kind where true premises guarantee a true conclusion — absolutely not. Experts are sometimes wrong; the conclusion could be false even if every premise is true. So formal logic, taking its ruler to the argument, stamps it a fallacy: the dreaded “appeal to authority.” And yet you reason this way a hundred times a day, sensibly, and you’d be paralyzed if you didn’t. Trusting a qualified expert’s testimony, provisionally, is not a mistake. So which is it — sound or fallacious?

The philosopher Douglas Walton spent a career showing that the question is wrongly posed, and the fix reorganizes how you can evaluate everyday reasoning. Most real arguments aren’t trying to be airtight proofs. They’re presumptive: they establish a conclusion provisionally, shifting the burden onto anyone who’d deny it, and they stand “unless” certain specific things turn out to be true. The expert argument isn’t broken logic; it’s a perfectly good presumptive argument that holds unless the expert isn’t really in that field, or has a conflict of interest, or is contradicted by the rest of the field. Those “unlesses” aren’t vague worries. For each type of argument they are a fixed, knowable list — and that list is the whole tool.

Walton’s move was to catalogue the types. He and his colleagues identified roughly sixty recurring argument schemes — appeal to expert opinion, argument from a sign, cause to effect, analogy, the slippery slope, ad hominem, practical reasoning, argument from ignorance, and on — and for each one wrote down its critical questions: the precise points at which an argument of that kind can be defeated. Appeal to expert opinion comes with six: Is the source a genuine expert in this domain? Did they actually say it? Is the claim within their field? Are they reliable and unconflicted? Do other experts agree? Can they produce the evidence? Argument from a sign asks a different set (could the sign have another cause? could it be faked?). The slippery slope asks whether each step really follows and whether the chain can be halted. The form tells you which questions to ask; you no longer have to invent the rebuttal from scratch.

This buys you two things people usually get wrong. First, it ends the false binary between “valid” and “fallacy.” The appeal to authority isn’t automatically a fallacy or automatically fine — it’s a legitimate scheme that becomes weak exactly when its critical questions go unanswered. The fallacy is the misuse of a good form, and now you can point to which question it dodged. Second, it sharply separates “defeated” from “unanswered.” If an expert clearly has a disqualifying conflict, a critical question is answered the bad way and the argument falls. But if you simply don’t yet know whether other experts agree, the argument isn’t refuted — it’s under-determined, still standing on presumption while you wait for more. Collapsing those two — declaring victory because a question is merely open — is one of the most common ways arguments get unfairly dismissed.

Which is why this turns out to be the natural backbone of auditing an argument for whether it holds together. If you want to test a piece of reasoning fairly — not knock down a caricature of it — the schemes tell you what kind of argument each step is, and therefore the specific questions it has to answer to stand; you press it at its real joints instead of flailing for a generic objection. And because each scheme separates a question answered the wrong way (the step is defeated) from a question simply left open (the step is merely unsupported), the audit can say exactly how an argument fails — or decline to call it failed when it is only under-determined.

Framework & implementation

This section uses Ora’s own terms for the parts of an analysis, so that if you open the actual mode and lens files they line up. Each is glossed in plain language on first use.

Pipeline execution

Walton’s schemes-and-critical-questions is one of the two required lenses of the Coherence Audit analysis — its lens_dependencies.required lists exactly toulmin-model and walton-schemes-and-critical-questions — and it is foundational: true in its own lens file, an argumentation-scheme-type catalog. Its applicability list names coherence-audit directly: an inferential audit can only test an everyday argument once it knows what type each step is, and the schemes supply the types. The mode runs at Gear 4, Ora’s most thorough setting — a Depth analyst and a Breadth analyst audit the reasoning in parallel, critique each other (cross-adversarial evaluation), and revise. Coherence Audit is conclusion-agnostic by design: it asks whether the argument as given establishes its conclusion, never whether the conclusion happens to be true.

Where the lens engages. It activates on its Detection Signals — a step that fits a recognizable presumptive form (an appeal to expert opinion, an argument from a sign, cause-to-effect, analogy, slippery slope). After the mode’s charitable reconstruction surfaces the argument’s strongest version and the Toulmin breakdown exposes each step’s claim, grounds, and connecting warrant, the schemes take over: the lens’s Application Steps identify the scheme each step instantiates by matching premise-and-conclusion patterns, list that scheme’s critical questions, and audit each one against the evidence — satisfied, defeated, or unaddressed.

What it contributes to the analysis. The per-scheme critical-question audit is a named section of the mode’s output — Walton scheme classification — where each step that instantiates a scheme carries its checklist with a per-question verdict. From there the lens feeds two more sections. A critical question answered the wrong way is what substantiates a named fallacy (the mode demands the four-part proof — quoted text, the inferential move, the violated principle, and why it fails here; a bare label never survives). And the supported-versus-open pattern feeds the per-move holds / fails / partially-holds verdict and, above all, the mode’s signature argument-wrong-vs-conclusion-wrong separation.

Cross-adversarial evaluation. At Gear 4 each analyst’s reading is critiqued by the other, which catches the lens’s signature failures, keyed to its Critical Questions and Common Failure Modes: scheme miscategorization (the wrong named scheme, importing an off-target checklist); strawman reconstruction (auditing a form the arguer never offered — which the mode’s charitable-reconstruction-first rule, its CQ1, exists to prevent); defeasibility collapse (treating an unaddressed critical question as a defeated one, claiming refutation where there’s only missing evidence); cumulative-case blindness (rating each scheme weak in isolation and missing the strength of their conjunction); and authority confusion (applying expert-opinion questions to a mere eyewitness, or vice versa). The evaluator presses the mode’s CQ4: that demonstrating an argument fallacious is not treated as showing its conclusion false (the fallacy-fallacy).

What the analysis will not do. It will not classify a step to a scheme without matching its actual premises; will not let an unanswered critical question masquerade as a refutation; and will not slide from “this argument fails” to “this conclusion is false” — coherence audit reports the conclusion as simply unsupported by this argument, a different and more honest finding.

Origin and evidence

The framework is Douglas Walton’s, the most influential modern systematization of informal logic. Argumentation Schemes (Walton, Chris Reed & Fabrizio Macagno, Cambridge University Press, 2008) is the canonical compendium — roughly sixty schemes, each with its premise pattern and its matched set of critical questions. It builds on Walton’s earlier Argumentation Schemes for Presumptive Reasoning (1996), which established the presumptive-reasoning frame (arguments that shift a burden of proof rather than compel a conclusion), and on A Pragmatic Theory of Fallacy (1995), which reconceived fallacies as the misuse of otherwise-legitimate dialogue moves rather than as fixed logical forms — the move that dissolves the “valid-or-fallacy” binary. Macagno and Walton’s Emotive Language in Argumentation (2014) extends the apparatus to value-laden and emotive language. The tradition descends from Stephen Toulmin’s model of argument (claim / grounds / warrant / backing / qualifier / rebuttal), which Walton’s schemes complement by classifying argument types where Toulmin analyzes a single argument’s anatomy. The schemes are now a backbone of computational argument-mining and AI argumentation systems precisely because they are explicit and enumerable.

Applications and common uses

The schemes are a working tool wherever reasoning needs to be classified and tested rather than merely read.

  • Argument and coherence auditing. The use that brings it here: reconstructing an argument charitably, then classifying each step’s scheme and running its critical-question checklist to find where the reasoning is supported, defeated, or merely unaddressed.
  • Steelmanning and charitable reconstruction. Rebuilding a position into a recognized, well-formed argument scheme so the strongest version is genuine, then locating the fair critique points via its critical questions.
  • Propaganda and rhetoric analysis. Naming the precise scheme a persuasive artifact leans on (appeal to popular opinion, argument from a sign, ad hominem) to surface its built-in defeasibility rather than just calling it “spin.”
  • Weighing competing hypotheses. Scoring how well each hypothesis is supported by its source-based and causal arguments, scheme by scheme.
  • Teaching critical thinking and building argument software. The schemes give a finite, explicit vocabulary for what makes everyday arguments strong or weak — usable by people and by machines.

In every case the payoff is the same: an argument is matched to its type, tested at the joints that type is known to fail, and judged on the right scale — presumptively supported, genuinely defeated, or honestly under-determined.

Failure modes and when not to use it

The lens’s characteristic ways of going wrong are catalogued in its Common Failure Modes:

  • Scheme miscategorization. Assigning the argument to the wrong named scheme, so the critical questions are off-target. The tell: the questions feel trivially satisfied or trivially unmet. Re-match the premise pattern carefully against adjacent schemes in the same family.
  • Strawman reconstruction. Auditing a reconstructed argument the arguer never made. The tell: the arguer wouldn’t recognize the form being tested. Quote the original premises before classifying; build the steelman first if the argument is fragmentary.
  • Defeasibility collapse. Treating an unaddressed critical question as a defeated one. The tell: the verdict claims refutation, but the questions only flag missing evidence on some conditions. Report “premise not satisfied” (defeat) and “evidence not yet supplied” (under-determination) separately.
  • Cumulative-case blindness. Auditing each scheme of a multi-scheme argument in isolation and missing the strength of their conjunction. The tell: every component rates weak, yet the whole feels strong. After the per-component audit, assess whether the components together raise the conclusion’s plausibility.
  • Authority confusion. Treating appeal to expert opinion and argument from position to know as interchangeable. The tell: domain-expertise questions aimed at a mere eyewitness. Separate expertise (knowledge of a field) from position-to-know (access to a particular fact) and apply the matched questions.

When not to reach for it. When the argument is a strict deductive proof (a mathematical or formal derivation), the presumptive machinery is the wrong tool — those are judged by validity and soundness, not by defeasible critical questions. When there is no actual argument to classify — a bare assertion, a mood, a vibe — there is no scheme to match, and forcing one manufactures structure that isn’t there. And the lens classifies and tests; it does not, by itself, settle which arguments are true — answering the critical questions still takes the underlying evidence, which is a separate job.

  • Coherence Audit — the analysis this lens is a required tool of; reconstructs an argument charitably, breaks it into Toulmin parts, and runs the Walton critical-question checklist on each step to see whether the reasoning holds.
  • Toulmin Model — the co-required lens: where Toulmin lays out a single argument’s anatomy (claim, grounds, warrant, backing, qualifier, rebuttal), the schemes classify what type each step is and supply its checklist.
  • Steelman Construction — another analysis that loads these schemes: there they keep a charitable reconstruction honest, ensuring the strengthened argument is a recognized form rather than a strawman.
  • Falsifiability — the kindred idea: a scheme’s critical questions are the concrete, answerable conditions under which a presumptive argument would be given up.