nucleation

CLAIM-342: What keeps a judgement–observation loop from confirming itself is evidence whose arrival or scoring does not pass through the interpretation it tests; action is one source of such evidence, not a necessary one, and a loop that selects its own evidence can lock in with action available

8 min read

Properties14
StatusActive
Version3
Rolethesis
Warrantinterpretive
Tagscognition, probabilistic-modeling, psychopathology-and-treatment
Linedistributed-agency, pragmatic-transport
Rests onCLAIM-330 — In the coupled meaning–influence model linearized at its symmetric state, with each field stable alone and nonnegative couplings, the coupling destabilizes symmetry exactly when βγ > (1 − α)(1 − δ): only the loop can do it, and the direction that grows mixes meaning and authority, CLAIM-141 — Causal-loop consistency and sheaf contextuality are both local-to-global problems, but passing to distributions affects them oppositely: a loop with no deterministic solution has a distributional fixed point, while a model with no global section in its supports has no global distribution either
GroundsTHEORY-285 — In an illustrative active-inference simulation of post-traumatic context inference, run once per phenotype, a prior that one is in a harmful context, made overly precise by a run of harmful trials under the "vulnerable" learning and forgetting weights, makes exploratory sampling look uninformative, so the agent attenuates the periphery, never receives the evidence that would correct the prior, and the belief sustains itself; lowering precision makes the attenuation policy less vigorous without correcting the prior, and the policy returns when precision is restored, THEORY-324 — A system adapts by trial and error to an environment whose effect on its essential variables it does not know only if a second feedback, through those variables, changes the parameters of its primary feedback; and if that feedback changes them whenever, and for as long as, an essential variable is outside its limits, every equilibrium the whole can reach is an adapted one, LIT-1181 — Action and behavior: a free-energy formulation
SupportsCLAIM-402 — Predictive success of a category across contexts warrants treating it as a real effective kind only when the evidence in those contexts does not pass through the category's own use, or when the effect of that use is modelled: a category that speakers or institutions act on can produce the regularities that confirm it, and contexts that share its use are not independent, CLAIM-404 — An organization can be governed from outside without any of its decisions being made for it, through the evidence its own procedures receive or through the criteria by which they weigh evidence; the two routes are distinct and each is possible without the other, so informational and evaluative independence do not imply each other; and where the organization's role is to evaluate another that affects its evidence or resources, the interpretation–authority loop can close between the two institutions, so that formal independence of the evaluator's procedure does not make its evaluation independent, CLAIM-413 — Grounding is not a property of an experience but a relation among an interpretation, what it purports to be about, and a process that constrains it independently; its current, developmental and epistemic senses come apart from one another, and in the current sense it comes apart from whether the interpretation occurs, so ungrounded is not unconstrained, and what makes an interpretive organization experiential need not be what makes its contents veridical
Objects toCLAIM-339 — A judgement–observation loop cut off from action loses one route for evidence it did not select, and the free-energy principle's reason for minimizing surprise, that a persisting system must keep its states in bounds, needs action on the world, though a loop can confirm itself with action available too
ComplementsCLAIM-341 — The loop that keeps a lucid agent's model in touch with what it models runs between judgement and observation: each judgement reshapes the observer whose later observations it then judges, and action on the world is not part of that loop, so active inference's action–perception cycle misplaces where the agent is active, CLAIM-338 — The selectivity active inference gives action, sampling what one expects to experience, is played in the coupled meaning–authority model by the authority space: influence decides which operators retrieve and so what the next judgement observes, which makes Friston's self-confirming sampling the model's authority-to-interpretation arrow and his agent a coupled loop run without independent scoring, CLAIM-357 — The frustration residual and governance regret separate structural infeasibility, realized violation and minimax suboptimality, and identify none of them as thwarting: a positive residual says no admissible update meets every declared requirement, a positive regret only that the actual update's worst residual exceeds the best achievable, which can hold while every requirement is met, and actual violation needs regret above the negative residual; aggregate performance does not tell infeasibility from realized violation, and the multipliers that certify infeasibility need not be the weights governance assigns, CLAIM-359 — A self-governing organization's robustness to compromised constituents is a property of the processes that allocate, audit and revise authority, not only of its constraints on harmful acts; safeguards against capture can also restrict the disagreement and revision that self-governance needs, and where they do security and self-governance trade off, while others, such as audits independent of the audited that protect dissent, serve both, so the two are distinct achievements and not opposed ones, CLAIM-427 — Evidence for the organizational-perspective framework comes in three grades: that different organizations produce the same ordinary behaviour, that they respond differently to independently specified interventions, and that a model representing the coupling of significance, authority and reproduction predicts those responses better than the strongest existing alternatives at matched complexity; only the third bears on the framework's defeater, and no grade counts if success is defined in the framework's own terms, CLAIM-446 — Some persistent anxiety may be governance-maintained rather than belief-maintained: anticipated threat grants defensive processes authority over action, attention and evidence acquisition; under the avoidance that authority selects, the hypotheses that the situation is dangerous and that it is safe predict the same observation, so repeated safety need not correct the threat estimate; relief reinforces the defense without being evidence of danger; and the regime can outlast a corrected belief, its pathology, where present, being defensive authority insensitive to corrective evidence at avoidable cost
Cited as “complements” byCLAIM-500 — The governance-induced kind of frustration is not idle: some failures of an organization's actual updates to meet demands that an admissible update would meet are explained by how influence is allocated, and not by step size, curvature, noise, conditioning or a fixed weighting rule
Cited as “illustrates” byCASE-186 — A proposed study: an organization in a world whose hidden regime reverses and then leaves its hypothesis family, with evidence generation, admission and scoring separately lesionable, its diagnostic machinery maintained by its own reciprocal production, from outside, or not at all, and finite fuel shared among operation, maintenance and diagnosis, to find whether a second correction after injury to the diagnostic and copying machinery separates arrangements that correct the first failure alike
Cited as “rests on” byARG-042 — The record's constructions of maintained correction show that self-maintaining organizations ought to preserve their corrigibility

Graph View

The claim

CLAIM-339 objects to CLAIM-341 that a judgement–observation loop with no action has a self-confirming minimum: judge so that what is observed confirms what is expected. The record already holds three results that bear on that minimum.

  • It exists for every loop. CLAIM-141: “Passing from values to distributions creates solutions.” A loop that feeds its outputs back as inputs has a self-consistent distribution whether or not it acts. What makes a fixed point wrong is a constraint from outside it that prescribes marginals, the kind a contextual empirical model imposes (TERM-042).
  • Action does not prevent it. In the active-inference model of THEORY-285, the agent can act, and still “as long as the sensory attenuation policy is selected, no amount of observations will render prior beliefs obsolete” (NOTE-766). The policy “withholds the only observations that could lower that precision”, and what it attenuates is a “sensory or cognitive periphery”. Lock-in comes from the loop selecting its own evidence, and action is one of the ways it selects. Friston et al. say so of action in general: it is “sampling selectively what one expects to experience”, and the weakness this brings “becomes the strength” (LIT-1181, NOTE-960).
  • Independent scoring prevents it. CLAIM-330 (Active, demonstrated): in the linearized operator model, capture starting at symmetry needs both arrows of CLAIM-297’s circuit, authority shaping interpretation and interpretation shaping endorsement. “Scoring operators independently of the dominant interpretation (Regime B, γ ≈ 0) removes the coupling-driven instability.”

So what anchors the loop is evidence whose arrival or scoring does not run through the interpretation it tests. Action on the world is one source of such evidence: it brings in what the agent did not ask for. Others are a source the agent does not select, such as an interlocutor, a scheduled probe or a body state, and a scoring rule fixed independently of the current interpretation. Ashby’s ultrastability has the same structure: a second channel, from the essential variables to the parameters of the first loop, that the first loop’s behaviour does not set (THEORY-324).

What it does not say

  • It does not say any amount of such evidence suffices. Where a loop pools an independent channel with one it selects, as the audience of Weatherall, O’Connor and Bruner does (THEORY-845, THEORY-855), the independent channel anchors it only as far as it outweighs the selected one.

  • It does not say an agent without action has independent evidence. The locked-in case asks whether one does (CASE-107). Damasio’s re-enacted body states (THEORY-500) and the visceromotor regulation that survives paralysis (NOTE-434) are candidates.

  • CLAIM-330 is a linearization at symmetry. It shows what independent scoring removes near that point, not that it prevents every form of lock-in.

  • It does not say what in a single mind plays the part of Regime B’s independent scoring. QUESTION-121 asks, with five candidates from the record: unselected input, the body, commitment before evidence, processing signals and other people. A paralysed person keeps the first, the second and the fifth inbound, so losing action need not mean losing the anchor.

Note of 2026-10-18: re-derived as independent audit

The exchange of 2026-10-17 proposed this claim’s thesis four times as a design requirement, without citing it.

  • A36 §4, on cryptographic commitments as “independent endorsement audits”: a commitment protocol lets an auditor check that a prediction “was not changed retroactively”, but “it only solves one problem: the timing and integrity of the prediction. It does not establish that the outcome is interpreted correctly, that the scoring rule is appropriate, or that the operator was exposed to representative evidence.” That is CLAIM-340’s limit: commitment fixes when the prediction is made, not how the evidence is read.
  • A39’s second case study: “endorsement based on independent performance evidence versus endorsement scored through the interpretations whose authority is being evaluated”, which A39 called “the best candidate for an early discriminating experiment”.
  • A40’s Study C, key control: “Evaluate the demands using evidence independent of the processes whose authority is being assessed.”
  • A41’s security table, whose row for independent audit reads “Prevent the actor being evaluated from controlling all the evidence of its competence”, and its variant that would “compromise the processes responsible for evaluating other processes’ competence. That tests whether the organization has any genuinely independent means of checking its own allocation of authority.” That variant is in CASE-108. It puts this claim under attack: the evidence that does not pass through the tested interpretation is itself scored by a process an insider may hold.

Note of 2026-10-21: re-derived three more times, and a performative instance

For public knowledge and institutions. A21’s criterion for media capture in the exchange of 2026-10-20, “whether the processes conferring epistemic authority can themselves be corrected by evidence independent of the authority they confer”, is this claim’s anchor stated for public knowledge. So is A20’s “A formally independent review procedure does not guarantee independent evaluation when the evidentiary standard itself is controlled by the interests being evaluated”. A22 added that the evidence channel may be “governed by what it is evaluating”. A23 §7 restated the claim again without citing it: “An expert may gain influence because their interpretations track independent evidence well. Alternatively, an interpretation may acquire apparent evidential support because it increasingly controls how evidence is selected and evaluated”, so “independent evaluation requires attention to the legibility and scoring of evidence, not merely formal separation between a decision-maker and its evaluator”. Its audit table listed “Governance can make its own evidence and authority self-reinforcing” as a new “Proposed institutional interpretation”. These are the second to fifth appearances. The institutional form is filed as CLAIM-404, which rests on this claim, with TERM-066 for capture.

A performative instance. A26 §2 built a toy in which a classification fixes the policy that generates its own evidence: θ_{t+1} = 1[P(Y = 1 | θ_t) > 0.5] with P(Y = 1 | θ) = 0.1 and 0.9. “Both θ = 0 and θ = 1 are stable fixed points”, each “consistent with the behavioral evidence generated under that classification” (checked). It is this claim’s loop with no independent evidence: every observation passes through the category under test. A26 tied it to Perdomo et al.’s performative prediction (LIT-1336) and Espeland and Sauder’s reactivity (LIT-1330), both registered unread then and read the same day (below), and drew this claim’s conclusion: “stability and accuracy under the prevailing intervention regime do not establish correctness under alternative regimes”. A27 carried it into QUESTION-028, filed as CLAIM-402, which rests on this claim.

Two readings of the same day. Weatherall, O’Connor and Bruner (NOTE-1038): their main remedy for a propagandist’s selective sharing is to hear more scientists, evidence that reaches the audience without passing through the propagandist (THEORY-855). Their audience’s state is one number, cumulative successes minus failures, fed by both channels (THEORY-845), so the independent channel helps exactly as far as it outweighs the selected one. Espeland and Sauder (NOTE-1050) report a loop whose evidence passes through the interpretation it tests: the reputation survey, in which raters score schools in part by the rank the scores feed (Stake 2006 as they cite it; Schmalbeck 2001 contra), the quantitative link cited, not shown (THEORY-854). Perdomo et al., read the same day, are in CLAIM-402.

Note of 2026-10-27: re-derived twice more, for an apparatus and for gluing

The exchange “Field Existence Question” reached this claim twice without citing it at first.

  • Observation. A19 §8 found it in the record and stated its condition in the exchange’s terms: “the world must have an opportunity to resist the inquiry’s specification”. It noted that this “need not mean an external conscious judge”. Independent sensory channels, perturbations, controlled interventions, competing models, or criteria the mechanism being evaluated cannot rewrite would all do (QUESTION-285).
  • Gluing. A21 §6 found a mathematical form of the self-confirming loop. An apparatus that revises its correspondence maps only to reduce disagreement can send them to zero, and then every observation is compatible and nothing is distinguished. The record already held the fact in THEORY-299: knowledge-sheaf consistency admits the all-zero embedding unless something contrastive is added. A21: “Self-consistency is not the same as correspondence with reality.” A22 §10 drew the test: whether correspondence learning “improves predictions on independently specified interventions”, the difference between “becoming more coherent and becoming more knowledgeable”. That is this claim’s evidence that does not pass through the interpretation it tests, made operational for one setting (CLAIM-471).

Note of 2026-10-27 (evening): the readings of the U15–U23 works

Three readings bear on this claim directly.

  • Discourse sheaves (NOTE-1222; THEORY-1088) are a formal self-confirming loop. The maps learn from disagreement among the agents’ own expressions, and nothing scores an expression against the opinion it expresses. The result converges to agreement the authors call “learning to lie”. The paper’s guards (minimal change, a conserved quantity) prevent collapse, not self-confirmation.
  • Robinson (NOTE-1221) shows the claim’s remedy in practice. Fusion always reaches consistency, so consistency shows nothing. The evidence that fusion helps is the error against a truth “withheld from the data presented to the sheaf model” (p. 25; Table 2).
  • Singer (NOTE-1224) checks his estimates against ground truth only in simulation. On real data the method’s only check is its own consistency.

Backlinks