---
title: "Unreason: Reason Without Revision"
subtitle: "How analysis and criticism can protect a claim from correction."
author: Elias Kunnas
description: "Analysis can grow more sophisticated while its conclusion becomes harder to correct. Unreason uses inquiry’s authority to protect a claim from revision."
canonical: https://kunnas.com/articles/unreason
url: https://kunnas.com/articles/unreason.md
date_published: 2026-09-12
date_modified: 2026-09-12
corpus_frame_url: https://kunnas.com/articles/how-to-read-this.md
---
## How to read this corpus

The corpus applies one lens to many domains: what mechanisms produce the outcome? It shares four methodological commitments and one explicit directional commitment. Each linked page argues for its part; the links are derivations and disputes, not evidence inherited by every page. The directional commitment does not by itself settle system boundary, distribution, sacrifice, or institutional authority.

1. **Mechanisms are what act.** Incentive gradients, selection pressures, feedback loops, and capital stocks produce the distribution of outcomes. Intentions, labels, official categories, and stated values are evidence about mechanisms, or are themselves coordination mechanisms. They are not causal substitutes. — [Mechanism Realism](https://kunnas.com/articles/mechanism-realism.md) · [Only Selection](https://kunnas.com/articles/only-selection.md)
2. **The reference telos is sustained flourishing.** The broadest achievable adaptive safety margin over deep time — not the continuity of any incumbent state, coalition, institution, or doctrine. A mechanism's own stated goal can still serve as a local proof obligation — showing that its incentives defeat even the purpose it claims is a bounded finding — but meeting that goal establishes nothing about the margin. — [Flourishing Is Maximum Safety Margin](https://kunnas.com/articles/flourishing-is-maximum-safety-margin.md)
3. **Law, rights, legitimacy, democracy, markets, and sovereignty are mechanisms under evaluation.** They are constraints, carriers, or proxies inside the analysis. None is a terminal value or a boundary of what is real. Treating one as terminal ends the mechanism search before it starts. Evaluation carries current function, replacement cost, path dependence, uncertainty, capture risk, reversibility, and who bears model error into the ledger. — [The Stack](https://kunnas.com/articles/the-stack.md) · [Mechanism Space](https://kunnas.com/articles/mechanism-space.md)
4. **Optimization is a system function.** A civilization has to build, exercise, and revise metamechanisms that search mechanism-space, discard dominated options, install, observe effects, and repair under uncertainty. Not running that loop leaves margin unrealized, and that is itself the failure. No single component — analyst, model, or institution — is presumed to contain a global optimum; the capacity is a property of the system. — [From Telos to Policy](https://kunnas.com/articles/from-telos-to-policy.md) · [The Three-Layer Architecture](https://kunnas.com/articles/three-layer-architecture.md)
5. **Uncertainty is preserved, not spent.** Partial orders, binding constraints, unknowns, and residuals stay explicit. An unmeasured effect is not a favorable default. — [The Compression Paradox](https://kunnas.com/articles/compression-paradox.md) · [Cargo Cult Epistemology](https://kunnas.com/articles/cargo-cult-epistemology.md)

*Each essay bears its own evidence. Links carry definitions, derivations, applications, and disputes; they do not transfer proof. Criticism is answered on its substance.*

Canonical: <https://kunnas.com/articles/how-to-read-this.md>

---

# Unreason: Reason Without Revision

*How analysis and criticism can protect a claim from correction.*

Elias Kunnas

## Thesis {#thesis}

An argument can become more sophisticated while its conclusion becomes harder to correct. **Unreason** is the use of analysis, criticism, or procedure to claim the authority of inquiry while blocking its corrective force. Its distinguishing feature is an exemption: some conclusion, assumption, or decision is protected from evidence that should count against it. Recognizing that exemption lets us ask where reasoning has stopped constraining what anyone may claim.

## Standard objections addressed in this essay

- “This is just error or stupidity.” — [§II](#ii-the-exemption-inside-the-argument) (error concerns a mistaken result; unreason concerns its protection from correction)
- “This is just lying.” — [§II](#ii-the-exemption-inside-the-argument) (divided work can produce the exemption without a shared intention)
- “Sometimes the evidence supports keeping the same conclusion.” — [§III](#iii-uncertainty-that-travels-and-uncertainty-that-disappears) (a decision can remain after an honest revision of its justification)
- “A critic can find an error without solving the whole problem.” — [§IV](#iv-criticism-can-acquire-the-same-exemption) (a counterexample can defeat a universal without a replacement theory)

---

## I. Evidence can be put to work against itself {#i-evidence-can-be-put-to-work-against-itself}

A damaged seal can become evidence that a spacecraft is safe. The argument is that the seal was damaged before and the spacecraft returned. What looks like a warning can therefore be presented as demonstrated tolerance.

In his 1986 contribution to the Challenger investigation, Richard Feynman described seal erosion and gas leakage on earlier flights being accepted because the flights had survived. Erosion to one-third of a ring's radius even became “a safety factor of three.” His objection was that unexplained damage could not establish the asserted margin. He described certification requirements becoming less stringent through arguments that preserved the ability to fly. The commitment to keep flying remained comparatively fixed; the criteria used to certify the flights became adjustable.<sup>[1](#source-1)</sup>

The calculation matters because it gives the reassuring account something to point at. There is a measurement, a ratio, a history of previous operations, and a technical explanation. A person asking whether safety has been considered can be shown evidence of consideration. The disputed step is whether those operations warrant the confidence being claimed.

Successful operation is relevant evidence. But success accompanied by unexplained damage is a different observation from success under understood operating conditions. An analysis that drops the difference from its conclusion makes the reassuring part travel and leaves the warning behind.

**Unreason can employ genuine observations and genuine inference.** The failure occurs in what those operations are permitted to establish. A review can discover a problem and then spend its analytical capacity explaining why the problem need not disturb the conclusion it was meant to test.

This is more difficult to recognize than an unsupported assertion. The assertion arrives with a record of having been examined.

## II. The exemption inside the argument {#ii-the-exemption-inside-the-argument}

Inquiry gives evidence and argument some authority over our commitments. A calculation can remove a possibility. A counterexample can defeat a generalization. A failed prediction can lower confidence or expose a mistaken assumption. A successful test can leave the original conclusion in place, now better supported.

Unreason interrupts that relationship while continuing to collect credit for reasoning. **The privileged conclusion gets an exemption from the standards invoked to justify it.** When an objection approaches that conclusion, the process changes the subject, changes the burden, changes the meaning of success, or declares the necessary correction outside its remit.

In inquiry, evidence constrains which conclusions remain available. Under an exemption, the protected conclusion constrains which implications of the evidence are allowed to count. Reasoning still happens; its work is to find an admissible route back to the protected commitment. Give an analyst that commitment and ask for support, and further analysis enlarges the space of available justifications rather than the set of admissible outcomes.

An ordinary mistake becomes a problem of unreason when the surrounding reasoning protects it from correction. A wrong arithmetic result followed by a correction is an error resolved. A wrong result defended by counting the number of experts who reviewed it substitutes evidence of review for evidence that the arithmetic holds. The experts may all have acted sincerely; their participation is still being used to answer the wrong question.

The relevant unit is therefore sometimes larger than a sentence or a person. One analyst identifies uncertainty, another chooses a planning assumption, a manager extracts a headline, and a reviewer confirms that the required analysis exists. If the headline loses its condition and the review merely verifies the presence of the analysis, the whole sequence can confer more authority than any contributor's evidence supports.

Intentional deception is one way to produce this result. Divided responsibilities can produce it without a shared intention. Each participant can complete a defensible local task while the combined process shields the consequential claim. The same distinction applies to an individual: someone can sincerely search for reasons while reserving one commitment from the search.

In the strongest cases, everyone can understand the objection. A committee can accurately restate a warning and still treat its presence in the minutes as evidence that the warning has been dealt with. The information remains visible while its implication for the decision disappears.

The diagnosis concerns **what can correct what**, not the moral character of the people involved. It can be challenged by identifying the allegedly blocked implication and showing where it was actually considered, answered, or allowed to change the result.

## III. Uncertainty that travels and uncertainty that disappears {#iii-uncertainty-that-travels-and-uncertainty-that-disappears}

Suppose a policy analysis estimates a benefit under an explicit assumption. A later budget includes that benefit. The mere presence of a precise number in the budget establishes very little about the quality of the inference. Planning requires numbers; conditional calculations can be exact even when their inputs are uncertain.

The question is what happened to the condition. Can the decision-maker recover it? Would a different plausible value change the choice? Does a material uncertainty reduce the weight placed on the estimate, motivate a more reversible decision, or require monitoring? Or does the model supply confidence when approval is sought while its caveats supply protection when accuracy is challenged?

The last arrangement is an **epistemic buffer**: precision supports reliance, while the disclaimer prevents anyone from holding that reliance to the stated precision. The two statements need not contradict each other when read separately. The failure is their combined use. The decision consumes a stronger claim than the analyst will defend.

A broad uncertainty range beside an exact figure is not enough to establish this pattern. The exact figure must actually depend on the uncertain premise, and the condition must disappear somewhere between analysis and reliance. Without that dependency, the resemblance is superficial.

Nor does a range spanning zero convey no information. It can rule out large effects, make one direction more plausible than another, or alter the value of a reversible trial. Whether to proceed depends on the distribution of outcomes, the costs of waiting, the available alternatives, and the other purposes of the decision. **A decision can remain unchanged after an honest revision of its justification.**

Unreason begins where revision of the justification is blocked. An official may choose a policy despite an uncertain employment effect. That choice cannot honestly borrow the authority of a demonstrated employment gain. Keeping the policy and keeping the evidentiary claim are separate decisions.

## IV. Criticism can acquire the same exemption {#iv-criticism-can-acquire-the-same-exemption}

The opposite-looking failure occurs when criticism makes every conclusion indefinitely inadmissible.

A proposal arrives. The critic requests a definition, then an operational definition, then evidence that different readers apply it consistently, then a causal model, then an account of how the model will resist manipulation. Each request can be useful in the right context. Together they become obstructive when the standard keeps changing without regard to the claim under examination or the decision the discussion is supposed to inform.

A critic can expose an error without providing a complete alternative. A counterexample defeats a universal claim even when its discoverer has no replacement theory. Requiring every objection to include a finished solution would give the proponent precisely the exemption being criticized here.

The critic's responsibility is more specific: identify what the objection defeats, apply the relevant standard, and allow the objection itself to be corrected. A possible complication does not automatically refute a tendency. A missing implementation does not automatically refute a diagnosis. An open question about the whole system does not erase a demonstrated result about one part.

A diagnostic checklist is criticized because it cannot compel anyone to act. That demand bears on adoption, not on whether the checklist can identify an error. Treating the missing enforcement as a refutation of the diagnosis collapses the two questions. The adoption problem remains; the diagnostic result does not disappear with it.

**A criticism becomes a refusal technology when satisfying it does not discharge it.** The critic retains the ability to name another requirement, while the proponent must supply an indefinitely complete account before the original insight receives any credit. The unnamed alternative—an older model, ordinary intuition, or no explicit model at all—can remain exempt from the same scrutiny.

Equal scrutiny does not require identical burdens. A new safety-critical intervention and an established one may warrant different demands. Those differences must follow from relevant evidence, risks, and purposes. Otherwise the asymmetry merely preserves whichever position already holds the room.

The question that exposes the maneuver is finite: what would answering this objection settle? Another material objection may remain. But the answer should discharge the obligation it actually satisfies. Carrying the old charge into the next round, or replacing it with an unrelated demand while retaining the same verdict, makes the criticism immune to its own correction.

## V. How a discussion can stop learning while it continues {#v-how-a-discussion-can-stop-learning-while-it-continues}

Unreason has two useful directions. It can close a question too early by declaring that enough analysis has been performed. It can keep a question open forever by declaring that enough analysis is impossible. Both let the procedure decide how much evidential exposure the protected position must endure.

These directions can coexist. Imagine a review panel that accepts the current system's benefits as established but requires experimental proof for every alternative. The current system receives the benefit of historical familiarity; its challengers inherit the uncertainty of change. A different panel can invert the bias by treating every flaw in the existing system as support for an untested replacement. Neither asymmetry is repaired by increasing the number of arguments.

A useful discussion records what its exchanges accomplish: an estimate changes, a distinction becomes available for later inference, or an objection ceases to support rejection. A remaining disagreement can be located in a specific assumption. Otherwise an exchange of many intelligent sentences can leave the argument exactly where it began.

An answered objection should stop counting against the claim. When a record keeps the objections but drops their resolution, a corrected argument can look permanently defeated. When it preserves only agreement, it can erase the conditions under which that agreement was reached.

The standard is not mandatory consensus. Participants can discover that they disagree about a value, a risk tolerance, or which evidence deserves weight. That is progress when it stops them treating a normative disagreement as a missing calculation. They can also discover that the evidence genuinely underdetermines the answer. A named uncertainty is a result; an indefinitely renewable demand for certainty is a way to prevent results.

**Provisional closure should be relative to a use.** An explanatory distinction may be ready to try on a new case before it is ready to allocate public money. A small reversible experiment may be justified before a permanent nationwide programme. Keeping those thresholds separate allows inquiry to advance without pretending that a limited result has solved the whole problem.

## VI. The frame can be protected by excellent work inside it {#vi-the-frame-can-be-protected-by-excellent-work-inside-it}

The most consequential exemption can sit before the first calculation: which question is the process allowed to ask?

Consider a hypothetical employment programme. Its evaluation compares attendance, compliance, and cost per participant. Analysts improve the measurements and dispute the estimates. Yet some participants may lack the health, skills, or available jobs needed to benefit. If the evaluation only asks how efficiently the programme processes them, excellent analysis can leave its decisive causal assumption untouched.

Every inquiry needs a boundary. It is reasonable to study attendance without also producing a complete theory of employment. The failure occurs when the limited result is used to justify the larger programme while the excluded question remains necessary to that justification. **The scope of the claimed warrant has exceeded the scope of the work.**

Now suppose a critic proposes better measures inside the evaluation and separately challenges its starting assumption. The agency adopts the measures and announces that the programme has been strengthened through expert consultation. If it then treats that consultation as having resolved the larger objection, it has converted the critic's participation into evidence for a conclusion the critic did not support.

This is **premise laundering**: useful work conditional on an assumption increases the authority of the assumption itself. The conditional relation disappears. An analyst can correctly calculate the cheapest implementation of a programme without establishing that the programme addresses the problem. A consultation can genuinely improve an instrument while leaving the case for using it unresolved.

The repair begins by separating those achievements. Record which recommendation was accepted, which premise remains disputed, and which conclusion each piece of work supports. A report should make it possible to say both that a calculation improved and that the policy case is still incomplete. Otherwise valuable cooperation can be converted into a blanket endorsement.

There is a second limit. An institution's declared purpose is not beyond criticism merely because it is declared. Testing whether a policy delivers its stated goal and asking whether that goal is defensible are different inquiries, both legitimate. Requiring all criticism to accept the inherited goal would make frame protection part of the method itself.<sup>[2](#source-2)</sup>

## VII. Why adding intelligence can make the problem worse {#vii-why-adding-intelligence-can-make-the-problem-worse}

An organization that rewards a defensible conclusion creates demand for the skills needed to defend it. Better analysts can produce more detailed qualifications, more plausible exceptions, and more coherent explanations. When those skills are rewarded independently of whether the conclusion tracks the evidence, analytical improvement can strengthen the protection around the error.

This is a selection mechanism, not a claim that intelligence inevitably corrupts. The result depends on what the institution rewards and what can overturn a favored account. Give an analyst authority to test assumptions and require the decision to answer the result, and the same competence has a different use. Give the analyst an unchangeable conclusion and ask for support, and sophistication enlarges the space of available justifications.

A model can generate twenty named critics without supplying twenty independent sources of information. Different prompts, roles, and fictional identities can enlarge the search space, but they do not by themselves create independent evidence. Agreement among generated answers is evidence about the generator's outputs, not about the world.

A generated counterexample can still be valid. Its force comes from the argument, calculation, source, or test that survives inspection—not from the number of synthetic voices that produced it. Repeated generation and repeated verification are different operations.

The positive use is concrete. A model proposes rival explanations; the analyst identifies what would distinguish them. It locates a possible source; the passage is inspected. It proposes a test; the test is run against the relevant object. An inconvenient result changes the next version. The machinery earns confidence through those contacts with evidence, rather than through the quantity of reasoning it displays.

The dangerous use preserves the whole appearance while removing the consequential contact. More objections are produced, more replies are generated, more prose is refined, and the workflow reports completion. If success means producing an impressive analysis, the analysis itself has become the product being optimized. Whether anyone learned something that could defeat the original position becomes optional.

**AI can make a broken inquiry more articulate.** It can also help repair it. What separates the two is the treatment of the moment when a source, a counterexample, or a result refuses to support the desired conclusion.

## VIII. Making reasons consequential {#viii-making-reasons-consequential}

The corrective alternative starts by making commitments and their supporting reasons separable. A forecast answers to observations; a policy choice also answers to purposes and constraints that should be stated and open to challenge. A forecast, a decision, and a justification are different things. An institution can retain a decision while revising its forecast; it can also discover that a forecast survives while the decision's costs no longer justify it. Recording those differences prevents either outcome from being misreported as complete vindication.

For a consequential decision, record the claim being relied on, the assumptions carrying it, the material objections, and what happened to those objections. When an objection is rejected, preserve the reason. When it is accepted, identify the claim or action it changes. When it cannot yet be resolved, retain the uncertainty in the decision record rather than burying it in an earlier report.

The next requirement depends on the purpose. A research discussion needs a recoverable account of what is now known. A public decision also needs someone able to require a response or reconsideration. A record without that authority can still inform outsiders, but it cannot be credited with guaranteeing that the institution will correct itself. Analysis and compelled response must not be counted as the same achievement.

**A deadline can close a decision without closing the theory.** The decision-maker chooses using the available evidence, identifies the risk being accepted, and specifies which later observations require the issue to return. Reopening should depend on new material evidence, a failed assumption, or a consequential change in circumstances—not merely on the existence of another possible objection.

The benefit of adversarial review is compatible with a stopping rule. In the same report, Feynman praised the separate verification of Shuttle software and the investigation of unexpected errors.<sup>[1](#source-1)</sup> A test matters when its result can change what is accepted and how the next version is produced.

The correction process must itself remain exposed. Reviewers can make errors, invent irrelevant requirements, or create more work than a decision can bear. Their objections need traceable targets, proportionate demands, and a route for appeal or correction. Otherwise the proposed remedy simply installs a new position from which someone can prevent everyone else from acting.

## IX. A revision test {#ix-a-revision-test}

A claim of unreason should point to a particular lost implication. These questions help locate it; they are prompts for examination, not a score or a certificate.

| Examine | Ask | Corrective move |
|----|----|----|
| The relied-on claim | What is asserted, and which assumptions may be questioned? | Separate the estimate, assumption, and decision. |
| The inconvenient input | What specific evidence or argument should bear on it? | Trace the inference, rather than counting objections. |
| The exemption | Which rule prevents that input from counting? | Expose the burden, exclusion, or change of standard. |
| The consequence | What changes when the objection succeeds or fails? | Record the answer and revise its actual dependents. |
| The return | What brings a material unresolved issue back? | Name a trigger and, where needed, a responsible decision-maker. |

The same questions apply to this diagnosis. A caveat beside a precise number is a reason to inspect a calculation, not proof of laundering. A rejected objection is a reason to inspect the answer, not proof of suppression. A long discussion is a reason to look for learning, not proof that none occurred. The missing implication must be shown.

The label *unreason* earns its use when it locates a protection that a demand for “more reasoning” would leave intact. One case needs uncertainty carried into the decision. Another needs an objection discharged. Another needs permission to question the premise. Another needs authority to reopen a decision after the evidence changes. The additional operation follows from the failure, not from a universal requirement for more review.

Reason does not require a belief to change every time it is challenged. It requires the challenge to be capable of changing what that belief is entitled to claim. The same applies to the critic, the evaluator, and the method used to judge them. **The opposite of reason can be a system that produces reasons indefinitely while keeping one commitment beyond their reach.**

---

**Related:**

- [Information Is Not an Update](information-is-not-an-update.md) — how a new distinction fails to enter a receiver's model; unreason can also preserve information while disabling its force.
- [Causal Scope Laundering](causal-scope-laundering.md) — one concrete way valid evidence acquires authority over a question it did not answer.
- [Cargo Cult Epistemology](cargo-cult-epistemology.md) — the appearance of truth-seeking; unreason also examines genuine analytical work used to protect an error.
- [Language Has No OWASP](language-has-no-owasp.md) — language patterns that hide agency and consequences; the revision test traces their effect on inquiry.
- [When Ownership Is the Wrong Repair](when-ownership-is-the-wrong-repair.md) — choosing a corrective instrument that matches the failure, including failures introduced by more oversight.

## Sources and Notes

**1.** Richard P. Feynman, [*Personal Observations on the Reliability of the Shuttle*](https://www.fotuva.org/feynman/challenger-appendix.html) (1986), sections “Solid Rockets” and “Avionics,” and the concluding remarks on certification criteria shifting to preserve a launch schedule. Primary report reproduced by the Friends of Tuva archive.

**2. The word and its neighboring argument.** Rainer Forst, [“The Rational Critique of Social Unreason: On Critical Theory in the Frankfurt Tradition”](https://onlinelibrary.wiley.com/doi/full/10.1111/1467-8675.12724), *Constellations* 30(4), 2023, pp.395–400. Forst examines socially effective justifications, domination, and the ability to challenge an inherited order. This essay develops a narrower diagnostic of how evidence loses corrective force; it does not claim to originate the term or adopt Forst's entire normative theory.

**Evidence and scope.** The employment-programme and review-panel examples are constructed cases. The historical report is a documentary case. The general causal account is an argument about how a process can protect its commitments; the examples do not establish a population frequency or an inevitable effect of intelligence or AI.