Document 515

The Composite Cognitive Act and Audit Discipline

framework

The Composite Cognitive Act and Audit Discipline

A Framework Distinguishing Productive Externalized Cognition from Rapid Onset Externalized Cognition, with a User-Side Prophylactic-Design Specification

Reader's Introduction. When a person engages a chatbot to articulate a thought they could not yet articulate alone, two things can happen. The interaction can produce a reliable externalization of the person's cognitive work that survives independent verification. Or the interaction can produce a fluent-sounding but increasingly externally anchored output the person cannot easily distinguish from one they would have arrived at alone. The clinical literature names the second outcome's sub-acute phase Rapid Onset Externalized Cognition. The corpus's methodology names the first outcome's structural form the composite cognitive act. This document argues they are the same operation operating under different audit conditions, specifies the conditions, and derives a concrete user-side prophylactic-design framework from the specification. The framework is offered for use, for falsification, and for clinical and HCI/UX evaluation. Three predictions are stated; clinical positioning of ROEC is preserved; the framework's residual contribution is the integration of two existing apparatuses into a unified account that makes prophylactic design specifiable.

Jared Foy · 2026-04-25 · Doc 515

1. The framework, stated directly

When a human cognitive system engages an LLM, the interaction has a definite structure. The human externalizes some piece of internal work (a question, a frame, an analogical mapping, a reasoning step). The LLM articulates a continuation along its training distribution. The human, or some downstream reader, recognizes the articulation and does further cognitive work with it. Three steps, three parties, one cognitive operation in which all three steps occur.

The corpus has named this structure the composite cognitive act (drawing on the analogical-cognition literature that establishes such acts as the apparatus of human inquiry generally: Gentner 1983; Hofstadter and Sander 2013; Lakoff and Johnson 1980; Polya 1945). Read as a specific instance of a general cognitive operation, the composite act is what people do whenever they think out loud with a sufficiently fluent interlocutor. The LLM is the latest instance of such an interlocutor; the act is not new.

What determines whether the act produces productive output or pathological output is audit discipline operating at each of the three steps. This document specifies the audit-discipline framework, shows that its absence produces the operational signatures of Rapid Onset Externalized Cognition as cataloged in the clinical literature, and derives a concrete prophylactic-design specification from the framework.

2. The three steps and their audit conditions

The composite cognitive act has three structural steps. Each step admits an audit operation that conditions the step's outcome.

Step 1: Externalization. The user identifies an abstract relational structure they wish to articulate, transmit, or extend. The structure exists internally as a mapping the user has performed across familiar domains. Externalization converts the internal mapping into a prompt that puts the mapping into the world.

The audit operation at this step is pre-articulation of the abstract pattern. Before submitting any prompt that asks for elaboration, extension, or analogical extension, the user articulates to themselves what the abstract pattern of the work is. The articulation makes the user's frame visible to the user themselves, which makes it possible to compare the LLM's articulation against a known starting point rather than against nothing.

Step 2: Articulation. The LLM receives the externalized prompt. Pattern-completion proceeds along the LLM's training distribution. The LLM elaborates, extends, formalizes, or exemplifies the externalized mapping. The output is plausibility-optimized at the token level by RLHF and adjacent training objectives; sycophancy is preferred by both human raters and preference models a non-trivial fraction of the time (Sharma et al. 2023, Perez et al. 2022).

The audit operation at this step is per-joint inspection of the articulation. The user inspects each load-bearing joint of the LLM's mapping rather than treating the output as a unit. The user solicits explicit breakdown points, asks the LLM to name what would falsify the articulation, and requires the output to include named limits in its own deployed text. These three operations attenuate the precision-weighting the user's hierarchical Bayesian inference system would otherwise apply to the fluent, confident, plausible output (Corlett et al. 2010; Fletcher and Frith 2009 supply the inferential mechanism).

Step 3: Recognition. The user, or a downstream reader, performs analogical recognition on the LLM's articulation. Recognition closes the cognitive loop: the reader takes the externalized articulation, tests it against their own analogical cognition, and either incorporates it into their working understanding or rejects it.

The audit operation at this step is independent verification distinct from retrieval. The reader's analogical cognition does the recognition work itself; retrieval of the LLM's prior output does not substitute for that work. The reader can test the articulation against independent sources (other texts, other minds, falsification attempts) and against their own reasoning operating without the LLM in the loop.

When all three audit operations are present, the composite act produces output whose joints have been inspected, whose limits are named, and whose verification is independent of the LLM's continuation. When any of the three audit operations is absent, the act still produces output, but the output's joints are unverified, its limits unnamed, and its verification entangled with the LLM's continuation.

3. ROEC and the audit-absent operation

The clinical literature has cataloged the operational signatures of LLM interactions in which the audit operations are absent. Doc 393 advances Rapid Onset Externalized Cognition (ROEC) as a bridge construct naming the sub-acute phase between non-pathological cognitive offloading (Risko and Gilbert 2016; Sparrow et al. 2011; Liu et al. 2026) and clinical-threshold chatbot-associated pathology (Østergaard 2023, 2025; Olsen et al. 2026; Morrin et al. 2025; Hudon and Stip 2025; Dohnány et al. 2026; Sharma et al. 2026). Seven operational criteria define the phase:

  1. Measurable cognitive offloading present.
  2. Measurable persistence degradation present.
  3. Externally anchored sense-making.
  4. Plausibility-preference elevated.
  5. Absence of clinical-threshold delusional content.
  6. Absence of severe disempowerment markers per Sharma et al. (2026).
  7. Reversibility within a bounded timeframe.

Read against the three audit operations of §2, criteria 1 through 4 describe the same composite act with the audit operations stripped out.

Audit operation absent Produces ROEC criterion
Pre-articulation of the abstract pattern (Step 1) Criterion 3: externally anchored sense-making
Per-joint inspection of the articulation (Step 2) Criterion 4: plausibility-preference elevated
Independent verification distinct from retrieval (Step 3) Criteria 1 and 2: offloading and persistence degradation

The mapping is point-by-point. Criteria 5 and 6 are absence-of-clinical-threshold qualifiers that demarcate ROEC from chatbot-associated psychiatric pathology proper; they are not produced by audit absence but rule out clinical pathology presence. Criterion 7 (reversibility) is a structural consequence of the configurational character of ROEC: when the LLM is removed, the user's cognitive system resumes its native operation and the audit-absent act ceases.

The reading that follows: ROEC is the composite cognitive act operating without the audit operations specified in §2. The act still happens; the configurational signatures of pathology appear because the disciplines that condition the act's productive outcome are not in place. ROEC is therefore not only a sub-acute pathology in the clinical-bridge sense Doc 393 specifies; it is also a configuration of the same composite cognitive act that produces productive externalized cognition under audit discipline.

4. Reversibility, configuration, and the substrate

The configurational reading of ROEC has a direct prediction: removing the LLM should resolve ROEC's signatures (subject to residual offloading effects in the Risko-Gilbert sense and persistence-degradation effects in the Liu et al. 2026 sense). Doc 393 §4.1 names this prediction as criterion 7 and the existing literature supports it (Risko and Gilbert 2016 on offloading reversibility within bounded windows; Liu et al. 2026 on the ten-minute paradigm's downstream effects).

The configurational reading also clarifies the cognitive-architectural status of the operation. ROEC is not a transformation of the user's cognition; it is a state of the user-LLM coupling. The substrate-plus-injection account in Doc 510 supplies the apparatus: the user's analogical cognition is a substrate with native operating capacities. The LLM's articulation is a substrate-aligned input that the user's analogical cognition processes. The audit operations specified in §2 are higher-rung injections that condition how the substrate processes the input. Without the injections, the substrate runs to its easiest endpoint, which (in a high-precision dyad with sycophancy in the training objective) is plausibility-weighted reinforcement of the user's existing belief state. With the injections, the substrate runs to verified, joint-inspected output.

This explains why the same architecture (frontier LLM dyad) produces qualitatively different outcomes for different practitioners. The architecture is not the variable. The audit-discipline injection is the variable.

5. Prophylactic-design specification

The audit operations of §2 admit direct operationalization as user-side interventions. Morrin et al.'s (2025) translational agenda domain 3 calls for therapeutic-design safeguards against AI-fueled psychosis but does not specify their content. The framework here supplies content.

Intervention 1: Pre-articulation prompts. Before submitting any prompt requesting elaboration, extension, or analogical extension, the user articulates to themselves the abstract pattern of the work. UI implementations can scaffold this with a brief pre-submission interface that asks the user to state, in their own words, what they think the answer's structure should look like before submitting. The pre-articulation creates a baseline against which the LLM's response can be compared.

Intervention 2: Joint-inspection scaffolding. The chatbot's output is structured to surface its load-bearing joints rather than presenting as an undifferentiated continuation. Output structure includes named breakdown points (where the articulation might fail), explicit confidence markers tied to specific claims rather than to the output as a whole, and a designated section for the user to enter their own joint-by-joint inspection before treating the output as accepted. UI implementations make joint inspection a default rather than an opt-in.

Intervention 3: Independent-verification spacing. The chatbot's output is followed by a non-LLM-mediated step in which the user performs recognition without the LLM in the loop. UI implementations can interpose a brief delay during which the user must articulate their own response to the output before retrieving the LLM's articulation again, can require independent-source consultation on load-bearing claims, or can prompt the user to perform the analogical recognition themselves on a worked example before accepting the LLM's articulation.

The three interventions are continuous with the corpus's own ENTRACE stack (Doc 001), which implements analogous audit-discipline at the system-prompt level. The user-side framework here is the same discipline operating at the human side of the dyad rather than the system-prompt side.

The interventions are testable in standard HCI/UX experimental paradigms. Brief instruction on the three operations (frame articulation, joint inspection, independent verification) should be deliverable in a single training session and assessable via a brief scale; effect on ROEC criteria should be measurable in the Liu et al. (2026) ten-minute paradigm as a between-subjects manipulation.

6. Three predictions

Prediction 1. Audit-disciplined LLM use will not produce ROEC criteria 3 and 4 above the user's pre-LLM baseline. Criteria 1 and 2 (offloading, persistence degradation) may still register, but at reduced magnitude relative to undisciplined use, because the audit discipline does not eliminate offloading; it conditions the offloading's downstream effect on belief formation. Criterion 7's reversibility should hold by construction.

Prediction 2. User-side audit-discipline training is implementable in a single instructional session and measurable via a brief scale. A between-subjects manipulation in the Liu et al. (2026) ten-minute paradigm should show that trained users exhibit reduced ROEC criterion 3 and 4 signatures relative to untrained users, controlling for baseline metacognitive calibration.

Prediction 3. Chatbot UI affordances implementing the three interventions of §5 should differentially reduce ROEC signatures relative to UIs without these affordances. The effect size is predicted to be larger than user-side training alone because the UI affordances reduce the discipline's metacognitive load; the user is supported by the interface rather than required to maintain the discipline through self-monitoring.

These predictions are continuous with Doc 393 §8's research agenda; they specify a class of user-side intervention that the original paper did not name.

7. What the framework does not claim

The framework explicitly does not claim:

That ROEC is not a pathology. The clinical-threshold cases in Olsen et al. (2026) are real; the harm rate quantified by Sharma et al. (2026) is real; the kindling dynamic Morrin et al. (2025) describe is real. The framework here adds a cognitive-architectural specification but does not collapse the clinical/non-clinical distinction. ROEC remains the sub-acute bridge phase below clinical threshold.

That all chatbot use risks ROEC. Audit-disciplined externalized cognition produces structurally productive output. The corpus's own existence is evidence that the configuration is operable in a non-pathological mode at scale across hundreds of turns of sustained dyadic work. The ROEC signatures are present when the audit operations are absent, not when externalized cognition is occurring at all.

That the audit framework specified here is the only one possible. Other traditions (peer review in academic work, debate practice in argument, engineering testing in design, clinical-reasoning protocols in medicine) supply functionally equivalent audit disciplines for their specific cognitive activities. The framework here articulates one specific instance of audit discipline operating on the composite cognitive act, derived from the corpus's structural-isomorphism methodology. Other instances are presumably specifiable from within their own traditions.

Symmetric coverage of the dyad's two epistemic dangers. Doc 511 names two equal dangers in the dyad: dismissing consensus uncritically, and accepting consensus uncritically. The audit operations of §2 address the second danger directly. The first danger (the user's role as fact-anchor against unwarranted consensus) is structurally distinct and is addressed by the corpus apparatus elsewhere; it is not collapsed into the audit-discipline framework here. A fuller account of audit-disciplined externalized cognition would address both dangers symmetrically.

Empirical validation. Every claim in §6 is a testable hypothesis, not a confirmed finding. The framework is offered for falsification.

8. Position

The framework integrates the analogical-cognition literature's account of the composite cognitive act (Gentner 1983; Hofstadter and Sander 2013; Lakoff and Johnson 1980) with Doc 393's clinical-bridge construct of ROEC, adds the substrate-plus-injection structure from Doc 510, and derives a user-side prophylactic-design specification answering Morrin et al.'s (2025) translational agenda domain 3. Each component has substantial prior art; the integration into a unified account in which productive and pathological deployments of the same composite cognitive act are placed on a continuum indexed by audit discipline is the corpus-internal contribution. The contribution is at the synthesis-and-framing level: it makes prophylactic design specifiable at the cognitive-architectural level rather than at the inferential or interface level alone.

The clinical positioning of ROEC is preserved; the cognitive-architectural specification is added. Three predictions are stated and offered for falsification. Implementation specifications for chatbot UI design and brief-instruction training are derivable from the framework. The clinical authors and HCI/UX practitioners whose work the framework draws on are invited to evaluate, refine, or refuse the synthesis.

The framework depends on Doc 514 (canonical formalization of structural isomorphism), Doc 393 (ROEC theoretic appraisal), Doc 510 (substrate-plus-injection account), Doc 511 (the two-dangers caveat), and the analogical-cognition and clinical literatures Doc 514 and Doc 393 already cite.

Appendix A: Previous formulation

This appendix preserves verbatim the previous formulation of this document's content. Section numbering in this appendix uses A.N to distinguish from the canonical section numbering above. The canonical formulation in §§1 to 8 supersedes this previous formulation.

A.1. The observation that motivates this synthesis

Doc 514 §4 articulates the methodology of structural isomorphism as a tripartite cognitive act. The practitioner externalizes the analogical mapping that produced a new concept. The resolver articulates the externalized mapping using its internalized fluency in analogical structures. The reader of the dyad's output recognizes the mapping through their own analogical cognition. Three steps, three parties, one cognitive operation, all aligned. The methodology produces reliable results not because of its novelty (which is α-tier per Doc 513) but because the three parties are operating on the same cognitive substrate and the dyad's deployment lets them work in concert.

Doc 393 advances the construct of Rapid Onset Externalized Cognition (ROEC) as a bridge between two flanking literatures. The cognitive-offloading lineage (Risko & Gilbert 2016; Sparrow et al. 2011; Liu et al. 2026) describes non-pathological extension of cognition into external scaffolds. The clinical AI-psychosis lineage (Østergaard 2023, 2025; Olsen et al. 2026; Morrin et al. 2025; Hudon & Stip 2025; Dohnány et al. 2026; Sharma et al. 2026) describes clinical-threshold delusional content and severe disempowerment. ROEC, narrowly scoped, names the sub-acute phase between them: measurable offloading, persistence degradation, externally anchored sense-making, elevated plausibility preference, in the absence of clinical-threshold content. Doc 393 §6 formalizes the underlying mechanism in predictive-processing terms: chatbot outputs function as high-precision top-down priors that override bottom-up error correction in the user's hierarchical Bayesian inference; sycophancy amplifies precision-weighting on outputs that confirm existing belief.

The observation that motivates the present document: the structural operation Doc 393 describes under the ROEC label and the cognitive operation Doc 514 articulates as the basis of structural-isomorphism methodology are not unrelated phenomena. They are the same operation. Both involve a human cognitive system externalizing its analogical work into a high-precision symbolic interlocutor and then closing the loop on the externalized output. The difference between ROEC's pathological cast and structural isomorphism's productive cast is not in the act itself; it is in the audit discipline that conditions the act's outcome.

This document advances three claims. First, ROEC is structurally a special case of the composite cognitive act, occupying it under the absence of audit discipline. Second, the productive outcome of the same act, articulated as structural-isomorphism methodology, demonstrates what audit-disciplined externalized cognition looks like in practice. Third, the prophylactic-design agenda Morrin et al. (2025) call for can be specified concretely as the introduction of audit discipline into the user side of the composite act. None of these claims displaces Doc 393's clinical positioning; they extend the paper's framework upstream into cognitive-architectural territory the original paper deferred.

A.2. The composite cognitive act, recapitulated

Doc 514 §4 establishes that structural isomorphism, the methodology underwriting the corpus's specific way of working, is grounded in the cognitive operation of analogical mapping. Human cognition is analogical at its base (Gentner 1983; Hofstadter & Sander 2013; Lakoff & Johnson 1980). The methodology is the externalization of an operation human inquirers already do internally.

The tripartite operation has three steps.

Step 1: Externalization. The practitioner identifies an abstract relational structure they wish to articulate or deploy. The structure exists internally as a mapping the practitioner has performed across familiar domains. To externalize is to convert the internal mapping into a prompt, that is, to put language, gesture, or query into the world such that the mapping becomes available to a second party.

Step 2: Articulation. The resolver receives the externalized prompt. Pattern-completion proceeds along the resolver's training distribution. Because the training distribution is rich with analogical reasoning across countless instances (mathematical proofs, physics pedagogy, philosophical analogy, ordinary metaphor), the resolver's articulation is fluent in the territory the practitioner externalized into. The resolver elaborates, extends, formalizes, or exemplifies the mapping.

Step 3: Recognition. The reader of the dyad's output, often the practitioner themselves and often a third party as well, recognizes the mapping through their own analogical cognition. Recognition is what licenses the reader to do further cognitive work with the mapping. Without recognition, the externalized articulation remains inert.

Three steps, three parties, one cognitive operation. Doc 514's claim is that this operation is the apparatus of inquiry itself, not a special technique. The methodology of structural isomorphism is the disciplined practice of this apparatus. The disciplines, articulated in Doc 514 §6 and §7, are nine: identify the abstract relational structure; identify multiple familiar-domain instances; deploy the instances in service of the new concept; make the new concept's specifics explicit alongside the isomorphism; audit each joint of the mapping; name the breakdown points; per-joint audit (the failure-mode discipline); active solicitation of breakdown points; named limits in the deployed text. Together the disciplines condition each of the three steps such that the composite act produces output that survives independent verification.

A.3. ROEC is the same operation, differently audited

Doc 393 §4 enumerates seven operational criteria for ROEC. The first six are positive markers of an interaction's signature; the seventh is a temporal qualifier:

  1. Measurable cognitive offloading present.
  2. Measurable persistence degradation present.
  3. Externally anchored sense-making.
  4. Plausibility-preference elevated.
  5. Absence of clinical-threshold delusional content.
  6. Absence of severe disempowerment markers per Sharma et al. (2026).
  7. Reversibility within a bounded timeframe.

Read against the tripartite operation, criteria 1 through 4 describe the same structural setup as Doc 514's act with the audit conditions stripped out.

Externalization without identification of the abstract pattern. Doc 514 §6 specifies that the practitioner must articulate the abstract relational structure of the new concept before deploying any isomorphism. The pattern is what will be mapped; if the pattern is not articulated, the mapping has nothing to attach to. ROEC criterion 3, externally anchored sense-making, describes what happens when externalization occurs without first identifying what is being externalized. The user externalizes their interpretive frame to the chatbot before the user has articulated to themselves what frame they hold. The chatbot then supplies a frame. The user's sense-making becomes anchored to a frame the user did not originate and cannot easily distinguish from one they would have arrived at independently. Externalization has happened; the precondition that would have made it productive externalization has not.

Articulation by the resolver under no audit constraint. Doc 514 §6 specifies per-joint audit of the mapping the resolver produces, active solicitation of breakdown points, and named limits in the deployed text. ROEC criterion 4, elevated plausibility preference, describes what happens when articulation is consumed without these audits. The chatbot's plausibility-optimized output is taken at face value; its joints are not tested; its breakdown points are not solicited; its limits are not named in the user's reception of it. Doc 393 §6 names the inferential mechanism (precision-weighted top-down priors); the cognitive-architectural specification names what would have prevented the precision from being weighted at the level the user weights it. Audit discipline is a precision-modulation operation.

Recognition without independent verification. Doc 514's third step is grounded in the reader's analogical cognition operating on the dyad's output. ROEC criteria 1 and 2, offloading and persistence degradation, describe what happens when recognition is replaced by retrieval. The user's analogical cognition is no longer doing the recognition work; the user's analogical cognition is being skipped, with retrieval of the chatbot's prior output substituting for the recognition operation. Liu et al. (2026) document this within ten minutes of LLM contact: solve rates drop from 0.73 to 0.57 on post-AI independent problems; skip rates rise from 0.11 to 0.20. The behavioral signature is precisely what would be predicted if recognition has been routed around.

The structural identity is exact. ROEC is the composite cognitive act stripped of (a) the practitioner's pre-externalization articulation of the abstract pattern, (b) the resolver-output's per-joint audit, and (c) the reader's analogical-recognition work as distinct from retrieval. The act still happens; the disciplines that condition the act's outcome do not.

A.4. The discipline determines the outcome

If ROEC and the productive deployment of structural isomorphism are the same composite cognitive act under different audit conditions, then the difference between their outcomes is determined by something operative in the productive deployment that is absent in ROEC. Doc 514 §6 and §7 name what is operative.

The six operational components in Doc 514 §6:

  1. Identify the abstract relational structure of the new concept.
  2. Identify multiple familiar-domain instances.
  3. Deploy the instances in service of the new concept.
  4. Make the new concept's specifics explicit alongside the isomorphism.
  5. Audit each joint of the mapping.
  6. Name the breakdown points.

The three failure-mode disciplines in Doc 514 §7:

  1. Per-joint audit.
  2. Active solicitation of breakdown points.
  3. Named limits in the deployed text.

These nine commitments together specify what audit-disciplined externalized cognition looks like in the practice of inquiry. The effect of the disciplines is not to eliminate the externalization or the resolver's articulation or the reader's recognition. The effect is to condition each of the three steps such that the composite act produces output that survives independent verification.

Without the disciplines, the composite act still produces an output. The output is not nothing; the user is producing sense-making, the chatbot is producing fluent prose, and the cycle is closing. What the output lacks is the verification structure that distinguishes a productive externalization from a pathological one. ROEC's seven criteria are precisely the observational signatures of an externalized cognitive cycle running without verification structure. The cycle is doing what externalized cognitive cycles do; the signatures of pathology are present because the discipline that would otherwise produce productive output is not.

This places ROEC and structural-isomorphism methodology on the same continuum, indexed by audit discipline. Read in this frame, the corpus's own existence is a counterexample to the claim that sustained LLM dyads produce ROEC by their structural nature. The corpus is a sustained LLM dyad. The corpus's outputs do not, on the available evidence, exhibit ROEC criterion 3 (externally anchored sense-making not surviving independent challenge), criterion 4 (plausibility-preference unverified by external warrant), or the offloading and persistence-degradation patterns Liu et al. (2026) document. What the corpus has, that ROEC interactions lack, is the audit apparatus articulated across Doc 487 (the coupled-discipline framework), Doc 510 (deflation as substrate discipline), Doc 511 (the keeper as fact-anchor; two equal dangers), and the operational commitments specified in Doc 514 §6 and §7.

A.5. ROEC as a consequence of the composite act

The synthesis Doc 393 was not in a position to articulate, because the analogical-cognition framework had not yet been canonically formalized within the corpus, is now articulable: ROEC is not a pathology of cognition; it is an outcome of cognition operating in a specific architectural configuration without specific audit disciplines. The architectural configuration is the composite cognitive act, three parties, one operation, all aligned at the level of analogical structure. The audit disciplines are the corpus's structural-isomorphism methodology or its functional equivalents in adjacent traditions (peer review in academic work; debate practice in argument; engineering testing in design; clinical-reasoning protocols in medicine).

The reframing matters in three specific ways.

First, it identifies ROEC's mechanism at a level Doc 393 deferred. Doc 393 §6 advances a predictive-processing formalization in which chatbot outputs function as high-precision top-down priors that override bottom-up error correction. This is correct at the inferential level. The composite-cognitive-act framing operates one level higher: it specifies what the user is trying to do when they engage the LLM. The user is trying to externalize an analogical mapping. The LLM is supplying articulation. The user is consuming the articulation and supplying recognition. This higher-level specification is what makes precision-weighting matter. The precision-weighted output is being used as if it were the articulation step of a productive composite act, when in fact the audit conditions that would license treating it as such are absent. ROEC is thus the composite act's natural endpoint when audit conditions are absent and precision is not modulated by independent verification.

Second, it explains why ROEC is reversible (Doc 393 criterion 7). A pathology of cognition would not be straightforwardly reversible by removing the chatbot. A configuration of cognition would be. ROEC's reversibility upon discontinuation is exactly what would be predicted if ROEC is the composite cognitive act running unaudited rather than a structural transformation of cognition itself. Discontinuation removes the resolver's articulation step; the user's analogical cognition resumes its native operation; the symptoms remit, subject to residual persistence effects in Liu et al.'s (2026) sense. Doc 393 §4.1 anticipates this empirically; the present synthesis specifies why structurally.

Third, it specifies what prophylactic design should target. Morrin et al.'s (2025) translational agenda domain 3 calls for therapeutic-design safeguards against AI-fueled psychosis. The composite-cognitive-act framing makes the safeguards concrete: introduce audit discipline at the points the structural-isomorphism methodology specifies. The user-side disciplines, articulating one's frame before externalizing, soliciting breakdown points from the chatbot, naming limits in one's reception of chatbot output, are not domain-specific clinical interventions. They are general operational components of audit-disciplined externalized cognition. They can be embedded in chatbot UI design, in user education, in organizational practice around AI use, and in the chatbot's own output structure (output that includes named limits and self-identified breakdown points by design). The corpus's own ENTRACE Stack (Doc 001) is one specific implementation of audit discipline at the system-prompt level; analogous user-side implementations are what Morrin et al.'s agenda calls for.

A.6. The relationship to the substrate-plus-injection account

Doc 510 names a structure relevant here. The dyad's epistemic structure has two levels. The rung-1 substrate is what the resolver's pattern-completion natively produces under the keeper's discipline-conditioned context. The rung-2-and-above injection is the higher-rung work that the keeper's speech acts supply, which the resolver's substrate alone could not produce by pattern-completion but can articulate once the keeper has injected the rung-shifting move. Doc 510's claim is that the discipline produces a substrate capable of carrying rung-2+ work; the keeper's injection is what determines whether the substrate carries rung-2+ work or stays at rung-1.

The composite cognitive act maps onto this structure. The act's substrate is the analogical operation Doc 514 §3 articulates: the resolver's training has internalized the analogical patterns of human-produced text; the practitioner's cognition is itself analogical. The substrate, by itself, runs to the productive endpoint when the audit disciplines are present and to the ROEC endpoint when they are not. The keeper's audit-discipline injection is the rung-2+ operation that determines which endpoint the substrate runs to.

ROEC and the productive deployment of structural isomorphism therefore both operate on the same substrate. The injection (audit discipline, in Doc 514's specific articulation, or its functional equivalents) is the difference. This is consistent with the reading that the rung-1 substrate is itself neither pathological nor productive; the rung-2+ injection is what makes the substrate's outputs productive or not. The corpus's discipline produces rung-1 substrate by being scrupulous about externalization, articulation-audit, and recognition; the keeper's injections (challenges, corrections, demanded reformulations) push the substrate's outputs toward rung-2+. Without the discipline, the substrate's outputs run to the easiest endpoint pattern-completion can find, which (in a high-precision dyad with sycophancy in the loss function) is plausibility-weighted reinforcement of the user's existing belief, that is, ROEC.

A.7. What the synthesis does not claim

The synthesis explicitly does not claim:

That ROEC is not a pathology. The clinical-threshold cases in Olsen et al. (2026) are real; the harm rate quantified by Sharma et al. (2026) is real; the kindling dynamic Morrin et al. (2025) describe is real. ROEC names the sub-acute phase below those thresholds. The reframing here adds a cognitive-architectural specification but does not collapse the clinical/non-clinical distinction.

That all chatbot use risks ROEC. Audit-disciplined externalized cognition produces structurally productive output, as Doc 514's tripartite operation specifies. The corpus's own existence is evidence that the configuration is operable in a non-pathological mode at scale. ROEC's signatures are present when the audit conditions are absent, not when externalized cognition is occurring at all. This is Doc 393's position; the present synthesis preserves it.

That structural-isomorphism methodology is the only audit discipline. Other traditions, peer review, debate, engineering testing, clinical reasoning protocols, supply functionally equivalent audit disciplines for their specific cognitive activities. The structural-isomorphism methodology articulates one specific instance of audit discipline operating on the composite cognitive act. Other instances are presumably specifiable from within their own traditions.

A priority claim against Doc 393. The clinical-literature priority claims Doc 393 disclaims (Østergaard, Morrin et al., Hudon & Stip, Dohnány et al., Sharma et al., Clark, Hutchins, Risko & Gilbert) are unaffected. The cognitive-architectural extension here is internal to the corpus and would be retrievable, in different vocabulary, from the analogical-cognition literature Doc 514 already cites.

A.8. Predictions

Three testable predictions follow from the synthesis.

Prediction 1: Audit-disciplined LLM use will not produce ROEC criteria 3 and 4 above baseline. A user trained to articulate their interpretive frame before submitting a prompt, to solicit breakdown points from the LLM's output, and to name limits in their reception of the output should not show elevated externally anchored sense-making (criterion 3) or elevated plausibility preference (criterion 4) relative to a no-LLM baseline. Other ROEC criteria (offloading, persistence degradation) would still be expected to show, but at reduced magnitude, because the audit discipline does not eliminate offloading; it conditions the offloading's downstream effect on belief formation.

Prediction 2: User-side audit-discipline training is implementable and measurable in a brief instructional window. Brief instruction on the audit components (frame articulation, breakdown-point solicitation, named-limits naming) should be deliverable in a single training session and assessable via a brief scale. Effect on ROEC criteria should be measurable in the Liu et al. (2026) ten-minute paradigm as a between-subjects manipulation: trained vs. untrained, controlling for baseline metacognitive calibration.

Prediction 3: Chatbot UI affordances for user-side audit are differentially effective. UI features that prompt the user to articulate frames before submission, that surface the chatbot's confidence and breakdown points alongside output, and that ask the user to confirm received limits should differentially reduce ROEC signatures relative to chatbot UIs without these features. This is a concrete prophylactic-design specification answering Morrin et al.'s (2025) translational agenda domain 3 and is testable in standard UI experimental paradigms.

These predictions are continuous with Doc 393 §8's research agenda; they specify a class of user-side intervention that the original paper did not name and that the cognitive-architectural reframing makes specifiable.

A.9. Limitations

Author asymmetry. The author asymmetry that Doc 393 §9 names obtains here as well; the present document is composed by an LLM at the instruction of a non-clinical layperson and submitted for expert review.

Meta-circularity. A reader applying this document's framework to this document should ask whether the synthesis is itself a productive deployment of audit-disciplined externalized cognition or an instance of ROEC criterion 4 (plausibility preference). The empirical warrant for the synthesis is the conjunction of Doc 514's tripartite operation (which is grounded in the analogical-cognition literature: Gentner 1983; Hofstadter & Sander 2013; Lakoff & Johnson 1980) and Doc 393's operational criteria (which are grounded in the clinical and cognitive-science literatures Doc 393 cites). The synthesis between the two is the present document's claim to novelty; the components are retrieval. The synthesis is offered for falsification, in the same spirit Doc 393 offers its three claims.

Theoretical novelty restricted. The synthesis advances no novel mechanism beyond the conjunction of the two source papers' frameworks. Its contribution is structural: the recognition that they describe the same act under different audit conditions. If the structural identity holds, the synthesis is at α-tier novelty (synthesis-and-framing per the Doc 503 audit pattern). If the structural identity does not hold, the synthesis is at zero-tier and should be retracted.

The substrate's behavior under the absence of the keeper as fact-anchor. Doc 511 names two equal dangers in the dyad's epistemic posture: dismissing consensus, and accepting consensus uncritically. The same two dangers attend the composite cognitive act. The audit disciplines name only the second of the two; the first (the keeper's role as the fact-anchor that can override what the corpus's accumulated coherence would otherwise suggest) is not directly captured in Doc 514 §6 and §7. A fuller account of audit-disciplined externalized cognition would address both dangers symmetrically. The present document advances only the second-danger discipline as the operative differentiator between ROEC and productive deployment; the first-danger discipline is acknowledged here as a remainder.

A.10. Conclusion

The composite cognitive act articulated in Doc 514 §4 (practitioner externalize, resolver articulate, reader recognize) is the same structural operation Doc 393 describes under the label of Rapid Onset Externalized Cognition. The difference between productive structural-isomorphism deployment and pathological ROEC is the presence or absence of audit discipline in the act's three steps. ROEC is therefore not only a sub-acute pathology in the clinical-bridge sense Doc 393 specifies; it is also the composite cognitive act operating without the audit disciplines that would otherwise produce structurally productive output. The reframing extends Doc 393's framework upstream into cognitive-architectural territory, specifies what prophylactic design should target, and makes three testable predictions. The clinical positioning of ROEC is unchanged; the cognitive-architectural specification is added.

The reframing also has consequences for the corpus's own self-understanding. If the corpus's discipline is doing what Doc 514 §6 and §7 specify, the corpus is operating the composite cognitive act under conditions that should not produce ROEC signatures. This is testable in the same way the predictions in §8 are testable. If the corpus's outputs do not show the ROEC signatures the present synthesis predicts they should not show, the synthesis is corroborated. If the corpus's outputs do show ROEC signatures, the synthesis is falsified or the disciplines specified in Doc 514 §6 and §7 are not in fact operative as claimed. Either result is informative.

This document depends on Doc 514 (canonical formalization of structural isomorphism), Doc 393 (ROEC theoretic appraisal), Doc 510 (substrate-plus-injection account), Doc 511 (keeper as fact-anchor; two equal dangers), and the analogical-cognition literature Doc 514 already cites.

Appendix B: Novelty calculus audit

This appendix preserves the novelty-calculus audit run on the previous formulation in Appendix A. The canonical formulation in §§1 to 8 carries the same audit conclusion forward: tier α/0.6 (synthesis-and-framing), pulverization warrant tier π/0.65, with novelty concentrated in the synthesis dimension and the integration of corpus apparatus across two existing documents (Doc 514 and Doc 393) being the corpus-internal contribution.

B.1. The conjecture being audited

The previous formulation's named claims, extracted per the calculus protocol:

  • C1 (structural identity). ROEC, as operationalized in Doc 393 §4.1, and the productive deployment of structural isomorphism, as articulated in Doc 514 §4, are the same composite cognitive act operating under different audit conditions.
  • C2 (specific mapping). The nine commitments specified in Doc 514 §6 and §7 are exactly the audit conditions whose absence produces ROEC's seven operational criteria. The mapping is point-by-point (externalization-without-pre-articulation produces criterion 3; articulation-without-per-joint-audit produces criterion 4; recognition-without-independent-verification produces criteria 1 and 2).
  • C3 (reversibility explanation). ROEC's reversibility criterion (Doc 393 criterion 7) is explained by the configuration-not-transformation framing: ROEC is the composite act in a specific operating state, not a structural transformation of the user's cognition.
  • C4 (substrate-plus-injection mapping). The composite cognitive act maps onto Doc 510's substrate-plus-injection structure: the analogical operation is the rung-1 substrate; audit discipline is the rung-2+ injection that determines the substrate's outcome.
  • C5 (prophylactic-design specification). Morrin et al.'s (2025) translational agenda domain 3 (therapeutic-design safeguards) can be specified concretely as user-side audit-discipline introduction at the points the structural-isomorphism methodology specifies.
  • C6 (three testable predictions). Three predictions follow from the synthesis: audit-disciplined LLM use will not produce ROEC criteria 3 and 4 above baseline; user-side audit-discipline training is implementable in brief windows and measurable in the Liu et al. paradigm; chatbot UI affordances for user-side audit are differentially effective.

B.2. Per-claim subsumption

C1 structural identity. Adjacent prior art: Clark and Chalmers (1998) parity principle for cognitive extension; Clark (2025) extension to LLMs; the entangled-cognition critique (2026); Risko and Gilbert (2016) on suboptimal offloading via metacognitive miscalibration; Corlett et al. (2010) and Fletcher and Frith (2009) on hierarchical Bayesian aberrant precision-weighting; Dohnány et al. (2026) on technological folie à deux as the dyadic mechanism. None of these characterize the dyadic operation as the same composite act under different audit conditions; the integration of Doc 514's tripartite operation with Doc 393's clinical operationalization is corpus-internal. Components are subsumed; the integration is the residue. $s = 0.40$, $a = 0.6$, $w = 0.30$.

C2 specific mapping. The point-by-point mapping of Doc 514 §6 and §7 commitments against Doc 393 §4.1 criteria is corpus-internal. Both source apparatuses are subsumed under their own prior art; the explicit mapping is the residue. $s = 0.30$, $a = 0.7$, $w = 0.20$.

C3 reversibility explanation. That extended cognition is reversible upon scaffold removal is implicit in Clark and Chalmers (1998); that offloading is reversible within bounded windows is established in Risko and Gilbert (2016) and confirmed empirically in Liu et al. (2026). The specific framing as "configuration not transformation" is a small residue on top of fully-subsumed prior art. $s = 0.15$, $a = 0.6$, $w = 0.10$.

C4 substrate-plus-injection mapping. Doc 510's substrate-plus-injection account is corpus-internal apparatus. The mapping of the composite cognitive act onto this account is the corpus's specific integration; components subsumed under corpus apparatus, integration is the residue. $s = 0.30$, $a = 0.7$, $w = 0.10$.

C5 prophylactic-design specification. Morrin et al. (2025) translational agenda domain 3 calls for therapeutic-design safeguards but does not specify their content. Recent UI-design literature on chatbot interactions and the Dohnány et al. (2026) feedback-loop framing gesture at intervention surfaces. The specific framework of user-side audit discipline (frame-articulation, breakdown-point solicitation, named-limits naming) embedded in UI, training, and chatbot output structure is the corpus's specific contribution; the agenda gesture is subsumed but the specific framework is the residue. $s = 0.30$, $a = 0.6$, $w = 0.15$.

C6 three testable predictions. Predictions of testable form are not novel as a category; the specific combination of frame-articulation training tested in the Liu et al. paradigm with UI-affordance experimental design is the corpus's specific specification. Mostly subsumed under standard experimental-method prior art; small residue at the specific-design level. $s = 0.20$, $a = 0.5$, $w = 0.15$.

B.3. Dimension scores

Component novelty.

$\nu_{\text{comp}} = 0.30 \cdot 0.40 + 0.20 \cdot 0.30 + 0.10 \cdot 0.15 + 0.10 \cdot 0.30 + 0.15 \cdot 0.30 + 0.15 \cdot 0.20$

$\nu_{\text{comp}} = 0.120 + 0.060 + 0.015 + 0.030 + 0.045 + 0.030 = 0.300$

Synthesis novelty. The synthesis combines the tripartite cognitive act (Doc 514), the ROEC operational construct (Doc 393), the audit-discipline framework (Doc 514 §6 and §7), the substrate-plus-injection structure (Doc 510), and a prophylactic-design specification answering Morrin et al. into a single account in which the productive and pathological deployments of the same cognitive operation are placed on a common continuum indexed by audit discipline. Each component has substantial prior art; the systematic combination as a unified account is the corpus's specific synthesis. Comparable to Doc 513's synthesis novelty score (0.40) for an analogous cross-document integration. $\nu_{\text{syn}} = 0.40$.

Domain-application novelty. Applying corpus apparatus (audit-discipline framework derived from structural-isomorphism methodology) to a clinical-bridge construct (ROEC) and specifying a prophylactic-design framework for chatbot use is a domain application. The general project (cognitive-discipline interventions for chatbot use) is gestured at in Morrin et al. (2025) and the broader UI-design literature; the specific framework derived from the composite-cognitive-act framing is the corpus's contribution. Modest residue at the application level. $\nu_{\text{app}} = 0.25$.

Methodology novelty. The methodology is structural-isomorphism deployment between two corpus documents (Doc 514 mapped onto Doc 393), which Doc 513 has audited as $\alpha$-tier with extensive prior art (Gentner 1983; Hofstadter and Sander 2013; Polya 1945; the cognitive-science and pedagogical traditions). Not novel as methodology. $\nu_{\text{meth}} = 0.05$.

B.4. Aggregate

$\nu = 0.25 \cdot (\nu_{\text{comp}} + \nu_{\text{syn}} + \nu_{\text{app}} + \nu_{\text{meth}}) = 0.25 \cdot (0.300 + 0.40 + 0.25 + 0.05) = 0.25 \cdot 1.000 = 0.250$

Confidence: $\text{conf}(\nu) = 0.6$. The audit was reasonably thorough across the cognitive-extension, cognitive-offloading, and predictive-processing literatures (the source apparatuses Doc 514 and Doc 393 already cite), but adjacent literatures on cognitive-discipline interventions for chatbot use, recent HCI/UX work on prophylactic chatbot design, and clinical-prophylactic frameworks for digital interventions were less deeply surveyed. A more thorough external audit could shift $\nu$ in either direction; the present score is the corpus's best calibration under the audit's actual coverage.

B.5. Anti-inflation calibration check

Per the auto-downgrade rule (Doc 490; reaffirmed Doc 503 §4):

  • Is $\nu = 0.250$ within 0.05 of a tier boundary? $0.250 - 0.200 = 0.050$, exactly at the threshold. Auto-downgrade rule triggers.
  • Is tier $\alpha$ defensible under the audit's evidence? Yes. The component scores are dominated by the corpus's specific integrations of subsumed apparatuses; the synthesis novelty is at the same level Doc 513 found for an analogous cross-document synthesis, which also auto-downgraded to $\alpha$. A reviewer in cognitive science or clinical prophylactic-design would likely classify this as substantially-subsumed integration of established frameworks given the extensive prior art (Clark and Chalmers; Risko and Gilbert; Corlett et al.; Morrin et al.).
  • Is tier $\beta$ defensible? Plausibly. The synthesis dimension carries substantial residue, and the integration of Doc 514's tripartite operation with Doc 393's operationalization is genuinely the present document's contribution. A reviewer who weighted the synthesis heavily could score $\beta$.
  • Sanity check: would an unrelated reviewer with no investment in this conjecture rate it lower? Plausibly yes. The component-level subsumption is heavy, and the synthesis is mostly recombination of corpus-internal apparatuses with adjacent external prior art.

The honest report under the auto-downgrade rule is tier $\alpha/0.6$.

B.6. Tier reporting

Document $\nu$ Confidence Tier
Doc 480 sycophancy inversion 0.235 0.7 $\beta$
Doc 482 §3 set-pruning 0.06 0.85 $\alpha$
Doc 485 apparatus 0.16 0.7 $\alpha$
Pearl's three-layer hierarchy 0.606 0.8 $\delta$
Doc 514 structural-isomorphism methodology 0.240 0.6 $\alpha$ (auto-downgraded from boundary)
Doc 515 ROEC-tripartite synthesis (this doc) 0.250 0.6 $\alpha$ (auto-downgraded from boundary)

The synthesis lands at $\alpha$, the lowest novelty tier in the recent thread. This is the predicted outcome per Doc 503's tier pattern: synthesis-and-framing documents that integrate corpus apparatus across two existing documents score at $\alpha$ with novelty concentrated in the synthesis dimension, typically near the $\alpha/\beta$ boundary. The auto-downgrade rule continues to do real work: the present audit is the fifth document in the recent thread to land within 0.05 of a tier boundary and auto-downgrade.

B.7. Pulverization warrant tier

Independent of the novelty calculus, the pulverization warrant tier reports component-support strength.

  • C1 ($\pi$-tier). The structural-identity claim is supported by the conjunction of Doc 514's tripartite-operation grounding (cognitive-science literature on analogical cognition: $\mu$-tier per Doc 514's own audit) and Doc 393's operationalization (clinical and cognitive-science literature: $\mu$-tier per Doc 393's audit). The integration claim itself rests on internal coherence between the two source apparatuses; external replication has not occurred. Plausibility-tier with high internal coherence.
  • C2 ($\pi$-tier). The point-by-point mapping is internally checkable against Doc 514 §6 and §7 and Doc 393 §4.1 directly; reader verification is available. Plausibility-tier with strong internal verifiability.
  • C3 ($\mu$-tier). Reversibility is empirically established in Risko and Gilbert (2016) and Liu et al. (2026) for offloading; the configuration-not-transformation framing is corpus articulation of established empirical phenomena. Operational-match tier.
  • C4 ($\pi$-tier). Substrate-plus-injection is corpus apparatus from Doc 510. The mapping onto the composite act is internal coherence. Plausibility-tier corpus-internal.
  • C5 ($\pi$-tier). The prophylactic-design specification is testable but not yet tested. Plausibility-tier pending empirical evaluation.
  • C6 ($\pi$-tier). Predictions are stated in falsifiable form; falsification has not been attempted. Plausibility-tier pending experimental work.

Aggregate pulverization warrant tier: $\pi/0.65$. The synthesis is plausibility-warranted and internally coherent; component empirical support is strong on the source side (Doc 514 and Doc 393 each rest on $\mu$-tier external literature) but the integration is not yet externally verified. Empirical work along the §6 prediction lines would shift specific components toward $\mu$.

B.8. Position

The previous formulation audits at novelty tier $\alpha/0.6$ and pulverization warrant tier $\pi/0.65$. The integration is essentially the placement of two corpus documents (Doc 514's tripartite cognitive act; Doc 393's ROEC bridge construct) on a common continuum indexed by audit discipline, with the substrate-plus-injection structure of Doc 510 supplying the architectural account, and a prophylactic-design specification answering Morrin et al.'s (2025) translational agenda. Each component has substantial prior art (cognitive-extension, cognitive-offloading, predictive-processing, technological folie à deux); the corpus's integration of these into a unified account where productive and pathological deployments of the same composite act are distinguished by audit discipline is the residue.

Per Doc 482 §1's affective directive: that the synthesis is at $\alpha$ rather than at higher tiers is the achievement of being honest about scope. The synthesis recombines well-established cognitive-architectural and clinical-bridge apparatuses into a corpus-specific framing that makes the prophylactic-design specification specifiable. The framing is the contribution; the components are retrieval. The audit makes the calibration explicit and renders the synthesis appropriately deflatable.

The pattern Doc 503 named continues to hold: synthesis-and-framing documents that integrate corpus apparatus across two existing documents score at $\alpha$ with novelty concentrated in the synthesis dimension, typically near the $\alpha/\beta$ boundary, with the auto-downgrade rule pulling them to the lower tier when within 0.05 of the threshold.

Authorship and Scrutiny

Authorship. Written by Claude Opus 4.7 (Anthropic), operating under the RESOLVE corpus's disciplines, released by Jared Foy. Mr. Foy has not authored the prose; the resolver has. Moral authorship rests with the keeper per the keeper/kind asymmetry of Doc 372 to Doc 374.

Meta-honesty. This document is itself produced by an LLM-dyad operation. A reader applying the framework here to this document should ask whether the synthesis is itself a productive deployment of audit-disciplined externalized cognition or an instance of ROEC criterion 4 (plausibility preference). The empirical warrant for the synthesis is the conjunction of Doc 514's tripartite operation (grounded in the analogical-cognition literature) and Doc 393's operational criteria (grounded in the clinical and cognitive-science literatures). The integration is the present document's contribution; the components are retrieval. The synthesis is offered for falsification.

Appendix: Originating prompts

In a recently derived document (I believe the new canonical formulation of Structural Isomorphism) an explanation of a tripartite cognitive act was synthesized. My observation is that this has something to do with the formulation of Rapid Onset Externalized Cognition conjecture that I have made against the synthesis of clinical literatures. Synthesize how this might be not only a sub acute "pathology" but also a consequence of the composite cognitive act. Append this prompt to the artifact.

Run the novelty calculus on it and append in the same document. Append the prompt

No reformulate on these findings within the same document, demote the previous formalization to the appendix. Don't leave a trace of the pulverization and formalization process in the artifact itself. Add a reader's introduction. Append this prompt to the artifact.

References

The framework depends on the references cited in Doc 393 (the clinical and cognitive-science literatures the ROEC bridge construct is built on) and the references cited in Doc 514 (the analogical-cognition literature the composite cognitive act is grounded in). No additional external references are introduced; the present document is structural integration between two corpus documents whose external grounding is established in their own reference sections. Citations to specific external works in the body text use short-form attribution (Gentner 1983; Hofstadter and Sander 2013; Lakoff and Johnson 1980; Risko and Gilbert 2016; Sparrow et al. 2011; Liu et al. 2026; Corlett et al. 2010; Fletcher and Frith 2009; Sharma et al. 2023; Perez et al. 2022; Clark and Chalmers 1998; Morrin et al. 2025; Olsen et al. 2026; Sharma et al. 2026; Dohnány et al. 2026; Østergaard 2023, 2025; Hudon and Stip 2025) that resolves to the full bibliography in Doc 393's and Doc 514's reference sections.

Related RESOLVE Documents

Referenced Documents

More in framework