Poster showing the basic dichotomy in Western philosophy: Idealism, represented by Plato and classical forms, opposed to Materialism, represented by industrial power and mass political imagery. The poster links this ontological split to the Biological LLM, artificial LLM ontology, and the article “A Hidden Split in Formal Reason.”

A Hidden Split in Formal Reason

Formal Capacity Is Not Governance Reliability

Directional Utilisation, Ontological Convergence, and Corrective Authority in Human and Artificial Decision Systems

Abstract

Advanced formal capacity is commonly treated as evidence of generalised rationality, self-correction, and governance competence. This article rejects that inference. It distinguishes five non-equivalent stages: formal capacity, partial realisation, directional utilisation, corrective utilisation, and governance reliability. A person, discipline, institution, or artificial-intelligence system may perform sophisticated abstraction, prediction, optimisation, historical reconstruction, and explanation while remaining unable to expose the ontology, purpose, representation, metric, or authority structure directing those capacities to consequential revision.

The article reconstructs a multilevel mechanism. Formal capacity is realised through disciplinary practice, socialisation, representational media, institutional roles, incentives, and public ontologies. Administrative-managerial institutions then translate accepted categories into classifications, thresholds, permissions, and civil consequences. Governance failure emerges where criticism remains expressible but cannot reach operative correction authority. This failure is analysed through Decision Sovereignty, Correction Sovereignty, Public-Grammar Sovereignty, Affective-Jurisdiction Failure, Durable Correction Asymmetry, Corrective-Voice Attrition, and—only under stricter downstream conditions—CEP-consistent persistence.

The constructive response is formulated as a two-layer corrective principle. At the level of governance essence, epistemological priority requires every operative ontology to remain provisional, justifiable, and reopenable by decision-bearing evidence. At the level of intellectual genealogy, objective-critical reason supplies the normative authority to judge ends rather than merely optimise means. The two principles are related but not identical: objective reason is the normative source; epistemological priority is its corrective institutional constitution. Epistemological priority is treated as a non-substitutable condition of governance reliability, but not as a sufficient condition without standing, Correction Sovereignty, implementation, reversibility, restoration, and learning.

A comparative test of ten high-capacity public textual-intellectual profiles revises the article’s initial two-cluster hypothesis. The evidence does not support a stable opposition between boundary/mechanism reasoning and narrative/social-symbolic reasoning. Nor does it support the predicted association of the former with institutional non-conversion or the latter with ontogenesis projection. The cases instead support a multidimensional model comprising boundary discipline, mechanism specification, source–path discipline, developmental integration, social-symbolic reach, medium discipline, corrective symmetry, and institutional conversion. Duality of Innate Cognition is consequently retained only as a non-load-bearing upstream hypothesis.

The AI-governance translation is functional rather than anthropomorphic. Model capability, evaluation, self-critique, and nominal human oversight do not constitute governance unless justified evidence can alter deployment permission. The proposed LoopGuard-AI architecture operationalises this requirement through four auditable permission states: SHIP, RESTRICT, HOLD, and ROLLBACK.

The article claims object formation, conceptual reconstruction, a comparative research design, and a proposed governance architecture—not completed causal, biological, or product validation. Its central conclusion is that advanced intelligence becomes governance-relevant only where the categories, objectives, representations, authority relations, and permission states directing its use remain available to consequential, reversible, and pressure-resistant correction.

Keywords: formal capacity; directional utilisation; corrective utilisation; governance reliability; epistemological priority; objective reason; instrumental reason; two-layer corrective principle; scientific-legitimation interface; boundary discipline; mechanism specification; source–path discipline; developmental integration; social-symbolic reach; corrective symmetry; institutional conversion; public ontology; correction sovereignty; AI governance; task ontology; LoopGuard-AI.

Epistemic Status

This article presents a theoretical reconstruction, an operationalisable family of distinctions, a multidimensional comparative test, and a proposed AI-governance architecture. It does not claim biological validation of Duality of Innate Cognition, psychological diagnosis of public figures, universal causal proof of the integrated mechanism, or production validation of LoopGuard-AI. Public figures are analysed only through bounded public textual-intellectual profiles. Internal RATIUM.AI materials establish construct provenance and architectural continuity; they do not constitute independent external validation. The principal internal line of development includes the predecessor article, the ADM/CIV governance formulation, the AI Configuration Paradox, the Central Equilibrium Problem, the stable-governance-layer formulation, the Priority of Epistemology, and the correction-mechanism analysis (Dunavich 2026a, 2026b, 2026c, 2026d, 2026e, 2026f, 2026g).

Title clarification: The “hidden split” identified by this article is not a division between two natural kinds of reasoner or two stable cognitive clusters. The comparative test did not support that binary classification as its final result. The title refers instead to a structural separation within advanced formal reason itself: the separation between demonstrated capacity, directional utilisation, corrective utilisation, operative authority, and governance reliability.

Unless explicitly stated otherwise, the formal expressions used below are schematic representations of directional and functional relations. They do not by themselves specify causal magnitude, probability distribution, equilibrium proof, or validated measurement models.

Part I — The Problem

1. The False Unity of Formal Intelligence

Advanced formal intelligence is ordinarily represented as a unified capacity. A person who demonstrates mathematical abstraction, strategic sophistication, scientific reasoning, conceptual precision, or complex historical interpretation is expected to possess a generalised ability to examine evidence, identify error, revise assumptions, and govern decisions rationally. The same presumption is increasingly transferred to artificial-intelligence systems. A model that solves difficult problems, writes functional code, reconstructs arguments, plans multi-stage action, or performs well on demanding benchmarks is treated as moving toward general reliability.

This inference is invalid.

Formal capacity establishes that a reasoning operation is possible under some conditions. It does not establish that the capacity is realised evenly across domains, activated under identity or authority pressure, directed toward the relevant object, converted into criticism of the governing frame, or connected to an institutional mechanism capable of changing action. A system may calculate, infer, classify, narrate, optimise, and explain while remaining structurally incapable of revising the ontology, objective, metric, representation, or permission regime directing those activities.

The problem is therefore not the disappearance of reason. Reason may remain conspicuously active. It may generate more analysis, more documentation, more explanation, more evaluation, and more procedural activity than ever before. Its failure may become harder to detect precisely because the visible products of reasoning remain abundant.

A scientific institution may produce methodologically sophisticated work while treating its inherited object definition as fixed. A hospital may collect incident reports while preserving the workflow that generates the incidents. A corporation may conduct extensive red-team evaluations while release incentives remain untouched. A government may provide review procedures whose outcomes cannot change operative policy. An AI system may identify its own uncertainty, describe a serious risk, and produce a compelling critique while lacking any channel through which that critique can modify the action it is authorised to perform.

In each case, reasoning exists. What is absent is a reliable transition from reasoning to correction.

Horkheimer’s distinction between objective and instrumental reason supplies an important antecedent. Reason may retain formidable competence in selecting and coordinating means while losing or never acquiring authority to judge the ends those means serve (Horkheimer 1947). Yet the present problem is narrower and more operational. It concerns not only the philosophical status of ends but the architecture through which any justified criticism—of an end, model, category, workflow, representation, or authority relation—can alter the state of the system.

The article therefore distinguishes criticism from correction.

Criticism is the appearance, registration, articulation, or circulation of an objection. Correction is a consequential alteration of the system in response to a justified objection. A system may permit criticism while remaining functionally closed. It may publish reports, host committees, collect feedback, employ evaluators, disclose limitations, or generate self-critical language without allowing any of those events to change permission.

A formally open system permits objection. A correctively open system permits justified objection to change the operative state.

This yields the article’s primary puzzle:

How can a person, institution, or artificial system possess advanced formal capacity while remaining unreliable as a governance mechanism?

The puzzle cannot be solved by dividing agents into rational and irrational populations. Nor can it be solved by assigning formal reasoning to science and narrative reasoning to the humanities. The comparative evidence developed in Part V shows that mechanism, boundary control, developmental integration, representation, and social-symbolic analysis appear in different combinations across disciplines and public profiles.

The relevant distinction is not between those who reason and those who do not. It is among the stages through which reasoning must pass before it becomes governance-relevant:

Formal Capacity→Partial Realisation→Directional Utilisation→Corrective Utilisation→Governance Reliability

(1)

Each transition introduces a separate failure condition. None is guaranteed by the preceding stage. The false unity of formal intelligence arises when these transitions are collapsed: capability becomes presumed use; use becomes presumed self-correction; self-correction becomes presumed authority; and authority becomes presumed reliability.

2. Research Object, Scope, and Claim Discipline

2.1 Research object

The research object is:

the non-conversion of advanced formal capacity into stable, consequential, and pressure-resistant corrective authority.

This object is neither exclusively psychological nor exclusively institutional. It can appear at multiple levels: an individual may reason differently across domains; a public textual profile may display recurrent explanatory patterns; a professional role may constrain which questions can be asked; an institution may separate evidence production from permission authority; a public ontology may determine which failures remain cognitively visible; and a deployed AI system may generate valid criticism without possessing any route to alter its authorised action.

These levels must not be conflated.

No inference may move from an individual task, public textual profile, professional role, organisational pattern, or civilisational grammar to another level without an explicit linking design. Structural similarity licenses comparison; it does not establish identity, causation, or common origin.

The same restriction applies to AI systems. An output cannot be attributed immediately to “the model” without separating base-model capability, system instructions, fine-tuning, classifiers, retrieval, tools, routing, interface, evaluator, and deployment policy.

2.2 Unit of analysis

The minimum analytical unit is:

U={Sys,Cap,Dom,Gen,Med,Ont,Auth,Π},

(2)

where Sys denotes the reasoning system or actor; Cap the relevant formal capacity; Dom the domain; Gen the genre; Med the medium; Ont the governing ontology; Auth the authority structure; and Π the operative permission state.

The unit is intentionally composite. Formal reasoning cannot be evaluated reliably by inspecting a detached output. A mathematically valid analysis may be irrelevant to the decision problem. A historically sophisticated narrative may conceal uncertain causal transitions. An accurate model may be embedded in an illegitimate task ontology. A justified critique may lack authority. An institution may possess the correct evidence while presenting it through a medium that suppresses uncertainty or affected-party consequence.

2.3 Governing ontology

A governing ontology is not merely a philosophical belief about what exists. It is the operative reality-picture encoded in a decision regime. It determines which entities are recognised, which relations are treated as causal, which outcomes count as harms, which agents possess standing, which purposes are legitimate, which uncertainties matter, and which corrections appear possible.

A governing ontology may be explicit, as in a legal definition or technical specification. It may also be distributed across databases, categories, benchmarks, interfaces, professional conventions, institutional mandates, and public narratives. The ontology governs before any explicit decision because it determines what the decision is understood to be about.

2.4 Claim classes

The article distinguishes five claim classes:

Conceptual definitions specify the proposed analytical vocabulary.
Externally supported background claims rely on established empirical, historical, or theoretical literature.
Synthetic inferences connect bodies of literature or internal constructs in a way not established directly by any single source.
Open hypotheses identify mechanisms requiring independent empirical testing.
Normative governance claims state conditions the article argues should govern consequential decision systems.

These classes must not borrow authority from one another. A useful definition is not an empirical finding. A coherent synthesis is not causal proof. An empirical association is not a normative permission. A governance proposal is not validated merely because the problem it addresses is real.

2.5 Non-claims

The article does not claim that formal utilisation is determined by innate cognitive type; that the ten comparative profiles form two natural clusters; that scientific and humanistic reasoning instantiate opposed formal structures; that developmental integration is equivalent to ontogenesis projection; that social-symbolic reach weakens mechanism or boundary discipline; that boundary discipline inhibits institutional conversion; that public textual profiles reveal private psychology; that institutional persistence proves equilibrium; that uncertainty or community dependence causes cognitive deficiency; that affect is epistemically inferior to reason; that sacred value, identity, or positive intersubjective content independently causes war; that technology necessarily produces closure; that AI systems possess human-like socialisation or innate cognition; that Horkheimer and Adorno themselves formulated the priority of epistemology over ontology; that scientific theory generally functions as a legitimation interface; that collective ontogenesis projection has been established as a causal or biological mechanism; or that LoopGuard-AI is a validated product or established regulatory standard.

Without these restrictions, the article would reproduce the same error it diagnoses: a bounded analytical model would be promoted into an ontology exceeding its evidential warrant.

2.6 Structural analogy without ontological identity

The article compares human cognition, professional discourse, institutions, and AI governance. These objects are not identical. A person has biography, embodiment, affect, social identity, and moral status. An institution has rules, roles, budgets, authority, memory, and legal consequence. An AI model has learned parameters and computational operations. A deployed AI system includes additional organisational and technical layers. A civilisational narrative is not an agent at all.

Yet structurally comparable questions arise across these domains: Can error be detected? Can the relevant level of failure be identified? Who may interpret the evidence? Who may alter the operative state? Can the decision be reversed? Can the correction mechanism itself be corrected?

The analogy concerns function, not shared substance or origin.

3. Formal Capacity Is Not Governance Reliability

3.1 Formal capacity

Formal capacity is the ability to reason through abstract relations, systems, rules, hypotheses, possibilities, counterfactuals, models, and structured inference. In human development, Piagetian formal operations provide one historical formulation of this threshold. In institutions, formal capacity may appear through statistical analysis, legal reasoning, clinical inference, technical review, strategic planning, causal modelling, or structured historical reconstruction. In AI systems, it may appear through mathematical solution, code generation, symbolic manipulation, planning, argument reconstruction, or multi-step tool use.

Formal capacity is necessary for many demanding forms of correction. It is not sufficient.

3.2 Partial realisation

Partial realisation is the uneven activation of formal capacity across domains, contexts, roles, media, or authority conditions. Developmental and educational research has repeatedly challenged a simple interpretation of formal operations as uniformly expressed across tasks and domains and has reported substantial context-sensitive variation in performance (Piaget 1972; Kuhn 2008; Moore and Rubbo 2012).

These findings support only a narrow claim: formal capacity may be realised unevenly. They do not establish the social or institutional mechanism proposed in this article.

Partial realisation is not necessarily hypocrisy or bad faith. The relevant capacity may fail to activate because the domain is represented differently, the task is not recognised as formally analogous, professional incentives change, identity costs become salient, time pressure narrows the reasoning horizon, authority discourages reopening, or the medium conceals the relevant uncertainty.

3.3 Directional utilisation

Directional utilisation is the recurrent configuration through which formal capacity is applied to specified objects, domains, problems, media, and authority relations. Formal reason does not vary only by degree. It may vary by direction.

A reasoning process may emphasise conceptual boundary, mechanism, source, temporal development, representation, public meaning, social identity, institutional consequence, or strategic stability. These emphases are neither mutually exclusive nor naturally distributed into two kinds of thinker. The comparative analysis in Part V shows that strong boundary discipline may coexist with strong developmental integration; mechanism may coexist with symbolic interpretation; and social-historical analysis may preserve rigorous source control.

Directional utilisation is represented as a multidimensional profile:

Pi=(B,M,SP,DI,SSR,MR,CS,IC),

(3)

where B is boundary discipline; M mechanism specification; SP source–path discipline; DI developmental integration; SSR social-symbolic reach; MR medium and representational discipline; CS corrective symmetry; and IC institutional conversion.

The dimensions are independent:

SSR↑⇏B↓,DI↑⇏SP↓,B↑⇏IC↓.

The configuration may vary within the same person or institution. A profile is a distribution across domains, genres, roles, and time—not an essence.

3.4 Corrective utilisation

Corrective utilisation is the use of formal capacity to revise the frame, metric, ontology, representation, authority relation, or procedure directing the reasoning process itself. A system may possess formal capacity, realise it across several domains, and use it in sophisticated ways while remaining non-corrective. It may analyse every variable except the variable that authorises its own action.

Corrective utilisation requires reflexive reach. Reason must be capable of turning toward its object definition, admissibility criteria, causal model, harm ontology, objective function, benchmark, proxy, interface, institutional mandate, and correction mechanism.

This is stronger than self-criticism as language. A model may state that its objective is incomplete. An evaluator may document that a benchmark omits affected-party harm. A committee may report that an institution’s review procedure is structurally inaccessible. None of these constitutes correction unless the finding can alter the operative state.

The minimum correction chain is:

Detection→Interpretation→Authority→Decision→Implementation→Review→Reversal or Escalation.

(4)

Failure can occur at every transition.

3.5 Governance reliability

Governance reliability is the stable capacity of a decision regime to preserve consequential correction under the conditions most likely to suppress it. A single good decision is not governance reliability. A single corrected output is not governance reliability. A successful incident response is not governance reliability. A review process that works only when no powerful interest is threatened is not governance reliability.

The relevant test occurs under uncertainty, time pressure, authority conflict, evaluator disagreement, incomplete evidence, identity threat, professional prestige, commercial incentive, public controversy, sunk cost, irreversible action, and institutional self-protection.

The full non-implication structure is:

FC⇏PR,PR⇏DU,DU⇏CU,

CU⇏CS,CS⇏GR.

(5)

Formal capacity is the beginning of the governance problem, not its solution.

Part II — The Cognitive and Contextual Genealogy

4. Prior Explanatory Orientations

Formal utilisation occurs within a prior field of attention. Before a person or institution adopts an explicit doctrine, it may already be disposed to treat some explanatory objects as salient and others as secondary. It may search first for mechanism, limit, source, sequence, identity, meaning, representation, or institutional function.

This article calls these tendencies prior explanatory orientations. They are candidate, non-exclusive tendencies of explanatory attention—not established psychometric dimensions, neurological systems, inherited biological kinds, or exhaustive classifications of reason.

4.1 Developmental orientation

A developmental orientation organises explanation through formation, temporal sequence, differentiation, continuity, transition, emergence, integration, and historical intelligibility. Its strength is that it renders formation visible. It asks how an object came to possess its present structure and how earlier states are related to later ones.

This orientation is indispensable where the object genuinely develops: organisms, institutions, technologies, scientific programmes, languages, political orders, or cultural forms. The error begins only when developmental grammar is allowed to perform work not supplied by mechanism or evidence. Temporal order may be mistaken for generative explanation. Retrospective coherence may be mistaken for necessity. Historical succession may be treated as maturation. Increased complexity may be described as intrinsic progress.

4.2 Entropic-boundary orientation

An entropic-boundary orientation organises explanation through limit, constraint, uncertainty, dissipation, mechanism, reversibility, source, and claim restraint. Its strength is that it protects explanatory levels. It asks where a concept begins and ends; whether metaphor has become mechanism; whether path has replaced source; whether model has become ontology; whether public authority has replaced evidence; and whether reconstruction has been promoted into proof.

Its characteristic risk is not ignorance of meaning but reduction. Boundary discipline may protect justification while failing to incorporate the social, symbolic, historical, or affected-party dimensions necessary for correction.

4.3 Non-exclusivity

The orientations are neither exhaustive nor mutually exclusive:

DO↑⇏EBO↓.

(6)

A profile may combine strong developmental integration with strong source discipline. It may analyse historical formation while preserving uncertainty and rival hypotheses. It may pursue ontological extension while marking domain limits. It may reconstruct social meaning through specified institutional mechanisms.

4.4 DIC after the comparative test

Duality of Innate Cognition remains a speculative upstream hypothesis concerning whether some recurrent explanatory tendencies have pre-disciplinary roots. It is not used here to classify persons or explain the ten-profile matrix. Any future DIC theory must demonstrate incremental explanatory or predictive value beyond discipline, genre, role, socialisation, institution, medium, ideology, incentive, and historical context.

The ladder from formal capacity to governance reliability remains intact even if DIC is rejected.

5. Socialisation and Positive Intersubjective Content

Formal reasoning does not develop or operate in an empty conceptual environment. Individuals enter worlds already organised by language, authority, identity, norms, institutional roles, and shared accounts of reality. Berger and Luckmann’s sociology of knowledge provides a broad antecedent: socially produced meanings may become objectivated, institutionalised, and encountered by later participants as an external reality rather than as contingent historical products (Berger and Luckmann 1966).

The present article uses Positive Intersubjective Content, or PIC, to identify the orientation-bearing content through which shared social life becomes cognitively and institutionally actionable:

PIC={R,N,I,A,C},

(7)

where R is shared reality; N norm; I identity; A authority; and C coordination and cost-bearing capacity.

PIC is not a deficit variable. Human beings cannot act collectively without shared reality, norms, identity, authority, and coordination. The governance question concerns the conditions under which these structures remain corrigible.

A social order becomes epistemically dangerous not because it contains shared meaning, but because its constitutive content governs who may criticise, which evidence is admissible, what counts as harm, and whether criticism can alter authority or permission.

The article proposes a bounded reciprocal hypothesis:

PIC Dependence⇄Reduced Corrective Generalisation.

(8)

Dependence on constitutive social content may raise the anticipated cost of unrestricted criticism. Where criticism repeatedly fails to become correction, participants may rely more heavily on stable shared content as the remaining source of orientation. This is an open institutional and motivational hypothesis, not a theory of deficient cognition.

6. Formal Operations without Generalised Correction

Piagetian formal operations provide a useful threshold concept because they distinguish reasoning about immediate concrete objects from reasoning about possibilities, systems, propositions, hypotheses, relations, and counterfactual structures. Yet possession of formal-operational capacity does not imply that the same reasoning will be activated across every domain (Piaget 1972; Kuhn 2008; Moore and Rubbo 2012).

The present article uses formal operations only as a threshold. It does not infer personal developmental stage, intelligence rank, or general governance competence.

The relevant distinction is:

Availability≠Realisation≠Generalisation≠Correction.

(9)

Availability means the actor or system can perform the relevant formal operation under some conditions. Realisation means the operation becomes active in a specific task or domain. Generalisation means the operation transfers to a structurally analogous problem in another domain. Correction means the operation is directed toward the frame, authority, or decision structure governing the actor’s own practice.

A person may recognise model-boundary error in physics but not in political memory. A historian may identify retrospective narrative construction in national mythology but not in a favoured theoretical account. A company may detect statistical drift while failing to question the proxy defining success. An AI model may criticise a decision rule when asked explicitly but continue to execute the same rule because no authorised state transition follows.

The gap is not proof of two cognitive kinds. It may arise from domain knowledge, professional role, genre, incentive, identity, risk, authority, interface, or institutional expectation.

Advanced reasoning may also increase the ability to defend an inherited frame. A sophisticated actor can generate more distinctions, identify more supporting evidence, reinterpret anomalies, construct auxiliary hypotheses, and absorb criticism into a wider explanation. Formal capacity can strengthen correction, but it can also strengthen frame preservation.

7. The Two Cultures as Formal Canalisation

Snow’s Two Cultures thesis diagnosed a consequential division between scientific and literary-humanistic intellectual life (Snow 1959). The present article does not map that distinction onto two forms of cognition.

The comparative test developed later shows that humanistic reasoning can be strongly mechanistic; scientific reasoning can be developmental and ontologically ambitious; narrative can preserve source discipline; and boundary reasoning can convert into institutional reform.

The contemporary governance problem is differentiated authority. Professional disciplines structure attention by defining recognised objects, methods, evidence forms, and jurisdictional boundaries, although the degree and direction of this structuring vary across fields and institutions (Abbott 1988). No actor necessarily owns the complete transition from evidence to public ontology, political purpose, institutional permission, civil consequence, and correction.

The forward chain is:

Specialised Evidence→Disciplinary Interpretation→Public Representation→Political Purpose→Administrative Permission→Civil Consequence.

(10)

An orphaned inference occurs where a consequential transition is made but no actor owns responsibility for its justification. Examples include evidence becoming public ontology, public ontology becoming policy purpose, policy purpose becoming classification, classification becoming automated permission, and permission becoming irreversible consequence.

A common corrective culture requires a reverse path:

Civil Consequence→Affected-Party Evidence→Cross-Disciplinary Review→Representation Correction→Ontology Reopening→Permission Change.

(11)

The solution to the Two Cultures problem is not intellectual fusion. It is a common correction architecture capable of integrating independently varying strengths. The governing question is:

Who owns the inference between cultures, and who can reopen it when its consequences become defective?

Part III — Ontological Convergence

8. Repeated Games and Recursive Reproduction

Directional utilisation may begin as a contextual feature. A person adopts one explanatory style in one domain, a profession privileges one evidentiary object, or an organisation selects one operational interpretation under pressure. None of these events alone establishes a stable ontology. Stability emerges through repetition.

Repeated interaction changes more than the frequency of a decision. It changes expectations, professional incentives, compatibility requirements, institutional memory, switching costs, justificatory vocabulary, and the apparent normality of the operative state. A provisional classification may become a standard field in a database. A temporary workflow may become the basis of staff training. A disputed model may become the default benchmark against which all alternatives must be expressed. A public interpretation may become the grammar through which later evidence is admitted or excluded.

The process is recursive:

Xt→St→Ft→Ut→Ot→Et→Gt→St+1,

(12)

where X denotes prior and contextual inputs; S socialisation and institutional conditioning; F realised formal capacity; U directional utilisation; O public or governing ontology; E the epistemic admissibility regime; G governance translation; and S(t+1) renewed conditioning produced by the governed environment.

The model is not a deterministic cycle. Competing interpretations, external shocks, leadership changes, new evidence, legal interventions, and affected-party action may redirect the process. Its purpose is to show that the consequences of reasoning become inputs into later reasoning.

8.1 From local decision to reproduction structure

A single decision may be locally defensible. An administrator may act on the best available evidence. A scientist may use an accepted model because no better operational alternative exists. A deployment team may release a system because expected benefit exceeds known risk under the current metric.

When similar decisions recur, however, they may produce conditions that make future alternatives less accessible. The decision may allocate resources toward its own continuation, create specialists whose authority depends on the existing framework, generate data only inside accepted categories, train users to formulate problems through the current interface, and make revision appear increasingly costly.

Locally Defensible Action+Recursive Institutional Effects→Collectively Restricted Correction.

(13)

This relation is a synthetic hypothesis, not a universal law. It becomes plausible where local actions demonstrably alter later incentives, infrastructures, categories, or switching costs.

8.2 Repetition changes payoffs

Repeated use may create technical dependencies, reputational commitments, legal precedents, standardised documentation, professional identity, data comparability, and public expectations. A revision that would have been inexpensive at an early stage may threaten major organisational assets later.

The expected cost of reopening the ontology rises even where the ontology’s evidential basis has not improved:

Institutional Cost of Revision≢Epistemic Strength of Existing Ontology.

(14)

Path-dependence literature supports attention to sequencing, increasing returns, learning effects, coordination effects, sunk investment, and rising switching costs (Pierson 2000; Sydow, Schreyögg, and Koch 2009). It does not establish the article’s stronger closure or equilibrium claims.

8.3 Critique absorption

Institutions frequently respond to criticism without changing their governing state. They may commission studies, add explanatory language, establish advisory committees, permit limited dissent, document objections, or create exceptional procedures. These actions may be substantively valuable. They may also absorb criticism.

Critique absorption is an article-specific construct for cases in which processing an objection lowers pressure on the institution while leaving the operative permission structure unchanged. A review may produce new language but preserve the same threshold. A consultation may broaden participation while leaving the object definition fixed. An appeal may reverse one outcome while leaving the generating rule intact.

The construct requires evidence of a recurrent gap between the level at which criticism is registered and the level at which the defect is generated. It must not be inferred from the existence of a committee or from failure to satisfy every critic.

8.4 Persistence is not equilibrium

Repeated outcome does not establish strategic equilibrium. An institution may persist because of habit, legal mandate, path dependence, technical incompatibility, resource scarcity, information asymmetry, hierarchy, or lack of a feasible alternative.

Persistence≠Equilibrium

(15)

The analysis distinguishes four levels:

Repeated outcome: a similar outcome occurs more than once.
Institutionalised pattern: repeated outcomes are connected to a common workflow, rule, classifier, authority practice, metric, or public grammar.
Closure regime: the pattern is combined with durable asymmetry between the path creating or preserving the outcome and the path capable of correcting it.
CEP-consistent persistence: the closure regime is combined with actor-specific continuation incentives, meaningful deviation costs, shared expectations concerning others’ responses, and persistence across genuine correction opportunities.

Episode→Pattern→Closure→CEP-Consistent Persistence.

(16)

Higher levels incorporate lower-level evidence and add new requirements. They must never be assigned from rhetorical severity, one dramatic event, or persistence alone.

9. From Disciplinary Grammar to Public Ontology

A technical result is not yet a public ontology. A model may be accurate within a bounded domain. A historical interpretation may be well supported within a specified corpus. A legal category may be coherent within a particular statute. A clinical classifier may perform adequately for one population.

Public ontology emerges when these bounded claims become part of the general reality-picture governing institutions and public action. The transition passes through representation, authority, repetition, and institutional use:

Disciplinary Result→Representational Medium→Authority Transfer→Public Ontology→Permission.

(17)

9.1 Description, explanation, ontology, and permission

Four levels must be kept separate:

Description: what pattern, state, or event has been observed?
Explanation: what mechanism or causal structure accounts for it?
Ontology: what kinds of entities, relations, capacities, and limits are treated as real?
Permission: what may an institution, operator, or system do on the basis of the preceding claims?

Description≠Explanation≠Ontology≠Permission.

(18)

A statistically observed association does not automatically establish a causal mechanism. A successful mechanism model does not necessarily establish a complete ontology. An accepted ontology does not automatically authorise every institutional use derived from it. Each transition requires independent justification.

Where the legitimate authority of a bounded scientific result is used to carry warrant across these transitions, the article calls the mediating arrangement a scientific-legitimation interface. The term does not describe scientific theory as such. It describes an institutional bridge through which authority may move from a bounded claim to a wider public ontology and from that ontology to permission. The interface is legitimate only where each transfer—description to explanation, explanation to ontology, and ontology to permission—receives independent justification. The secondary metaphor of an “institutional API” is useful only if it preserves this distinction: the relevant object is the translation layer, not an allegation that the underlying science is false or intentionally designed to legitimate governance.

9.2 Source, generative capacity, path, and contingency

Building on—but extending beyond—Ben-Menahem’s treatment of contingency within causal constraint, this article proposes an explanatory-order distinction among source, generative capacity, path, and contingency (Ben-Menahem 1997, 2018):

Source→Generative Capacity→Path→Contingency

(19)

Source concerns the constitution of the system or possibility space. Generative capacity concerns what the constituted system can produce. Path concerns the route through which one possibility becomes realised. Contingency concerns sensitivity to conditions, branching, and the fact that alternative paths could have occurred.

This order is explanatory rather than necessarily chronological. Recursive outcomes may alter future possibility spaces. The claim is narrower: a path-level account cannot substitute automatically for an account of the system, architecture, or generative capacity that made the path possible.

9.3 Variation and architecture

A further compression occurs where observed variation is treated as sufficient explanation of organisation:

Variation≠Architecture.

(20)

Variation describes differences among states, components, or frequencies. Architecture concerns the relational organisation through which those components become a functioning whole. This distinction does not deny that accumulated variation can alter architecture. It requires the bridge to be specified through regulatory interaction, developmental pathway, constraint, selection, integration, material threshold, or system-level reorganisation.

9.4 Model limitation and ontological prohibition

Every model excludes information. Exclusion is often necessary. The error begins when a model’s inability to represent an object is treated as evidence that the object is unreal or irrelevant:

Model Omission⇏Ontological Absence.

(21)

The reverse error also occurs. The existence of an omitted phenomenon does not establish that it should govern the decision. Affected experience, cultural meaning, or moral value may be real without settling technical causation or public permission. Governance must integrate levels without allowing any level to claim universal jurisdiction.

9.5 Truth-value and public deployment

A proposition may be true while its public use is misleading or illegitimate:

Truth-Value≠Public Deployment

(22)

A bounded scientific result may be simplified, detached from its assumptions, repeated through authoritative institutions, embedded in education, and translated into public ontology. The result itself need not be false. The deployment may nevertheless exceed its warrant.

9.6 Canonization-Anomaly

The article uses Canonization-Anomaly to describe the enlargement of bounded claims into a stronger public truth-status. The process may include truth inflation, framing compression, authority transfer, salience anomaly, and context loss. Canonisation need not require conspiracy or deliberate deception. It may emerge through education, media, administrative convenience, disciplinary prestige, repeated citation, interface design, and institutional demand for clear decisions.

9.7 Medium as an epistemic layer

The representational medium is not neutral. The same conclusion may appear as a probability distribution, categorical label, narrative explanation, risk colour, ranking, warning, or automated action. Each representation changes what becomes salient.

Content+Medium+Authority→Public Effect.

(23)

The medium can preserve a claim’s boundaries or erase them. In AI systems, the interface may transform model output into recommendation, recommendation into default, default into action, and action into an apparently objective institutional record.

10. Two Ontological Attractors

The article introduces two article-specific regime-level ideal types. They are not person types, psychological diagnoses, empirical clusters, or completed taxonomies. They model different explanatory hierarchies: different ways of ordering development, entropy, source, architecture, variation, and generative sufficiency.

10.1 Attractor I — Developmentalised Materialism

The first attractor may be stated polemically as “matter is all-powerful.” The academically operative claim is narrower: material process is treated as possessing broad source-level generative sufficiency, so that once matter, variation, time, and interaction are granted, the major architecture of reality is presumed explainable through developmental history.

The governing hierarchy is:

Entropy⊂Development.

(24)

The attractor supports historical explanation, emergence, continuity across levels, resistance to static essentialism, and investigation of how complex organisation arises through natural processes. Its risks arise where path is promoted into source, temporal sequence substitutes for generative mechanism, possible accumulation becomes sufficient architecture, or retrospective coherence becomes ontological completion.

The error is not materialism as such. It is explanatory closure through unbounded developmental sufficiency.

10.2 Attractor II — Entropy-Bounded Development

The second attractor begins from a different hierarchy:

Development⊂Entropy.

(25)

Development is real but bounded. Organisms develop. Institutions form. Technologies change. Scientific theories evolve. Yet every developmental process operates inside material constraints, finite resources, information limits, irreversible losses, and prior organisational conditions.

The attractor emphasises source, architecture, boundary, constraint, level, and the difference between realised path and generative capacity. Its strengths include resistance to teleology, model restraint, attention to irreversibility, and protection against promoting process description into complete ontology. Its risks include reduction and non-conversion: strict boundary discipline may fail to explain social meaning, historical formation, institutional legitimacy, or the symbolic structures governing public action.

10.3 Analytic symmetry without substantive neutrality

The attractors are analysed symmetrically. That does not mean the article is substantively neutral between every proposition associated with them. The manuscript adopts a methodological and ontological commitment:

Development should be analysed as bounded within entropy, constraint, prior structure, and specified generative capacity.

This does not deny emergence or historical formation. It requires developmental explanation to identify the changing object, continuity conditions, mechanism of transition, possibility space, and boundary of the claim.

10.4 Profile, orientation, and attractor

Three constructs must remain distinct:

Orientation≠Profile≠Attractor.

(26)

An orientation is a possible tendency of explanatory attention. A profile is an observed distribution across domain, genre, role, period, and corpus. An attractor is a regime-level hierarchy of explanatory settlement.

High developmental integration does not establish Developmentalised Materialism. High boundary discipline does not establish Entropy-Bounded Development. Attractor assignment requires analysis of the governing explanatory hierarchy, not vocabulary or profile score.

11. Ontology as the Higher-Order Governance Object

Governance is ordinarily designed around decisions, risks, outputs, and procedures. The deeper object is ontology. Ontology determines what the system is deciding about before any rule is applied. It specifies the relevant entities, unit of harm, meaning of success, recognised agent, available correction, and boundary between legitimate and illegitimate evidence.

The article proposes the following conceptual relation:

Ontology is consequence-prior; epistemology is correction-prior.

Ontology determines the field of possible consequence. Epistemology determines whether that ontology remains answerable to justification.

11.1 Accepted ontology and permitted epistemology

In many regimes, the practical order is:

Accepted Ontology→Permitted Epistemology.

(27)

Once the institution accepts an ontology, it defines which questions count as relevant, which experts possess jurisdiction, which evidence formats are admissible, and which objections appear category errors. Epistemology then operates inside the accepted ontology rather than testing it.

11.2 The epistemological-exception hypothesis

The article uses epistemological exception as the name of a provisional historical hypothesis: that some institutional traditions have attempted to place accepted ontology downstream of public justification.

Epistemic Justification→Provisional Ontology.

(28)

The present manuscript does not establish the historical uniqueness, geographical exclusivity, continuity, or comparative superiority of that development. The governance significance lies only in the order of priority: ontology should remain downstream of justification, even where institutions require temporary operative closure.

11.3 Instrumental closure

Modern institutions may preserve the language of critical epistemology while reversing the order operationally. They retain peer review, audit, consultation, appeal, transparency, and evidence-based policy, yet the accepted ontology increasingly determines what those procedures are permitted to question.

Instrumental reason asks how to achieve the objective, reduce cost, improve prediction, increase compliance, or manage risk. Corrective reason asks whether the objective is legitimate, whether the proxy represents the object, whether the ontology excludes relevant harm, whether the authority is properly constituted, and whether the permission state should exist (Horkheimer 1947).

An institution may be excellent at the first and structurally weak at the second.

11.4 The Two-Layer Corrective Principle

The preceding diagnosis identifies a recurrent inversion: accepted ontology determines which epistemology is permitted to examine it. A corrective order must reverse that dependency without pretending that institutions can operate in a state of permanent ontological suspension. The article therefore proposes a two-layer corrective principle.

The first layer states the governance essence of the solution. The second identifies its intellectual genealogy. They are connected, but they are not interchangeable.

11.4.1 Governance essence: epistemological priority

Epistemological priority means that every operative ontology must remain provisional relative to the procedures by which it is justified, criticised, revised, and withdrawn. It does not mean that reasoning begins without assumptions, that ontology can be eliminated, or that institutions must reopen every premise before every action. It specifies an order of authority:

No accepted reality-picture may acquire final permission authority before the conditions of its justification and correction have been specified.

Ontology remains consequence-prior because it defines what the institution can see and do. Epistemology remains correction-prior because it determines whether the ontology is entitled to retain that consequence-producing position.

The principle therefore concerns the relation between knowledge and institutional force. An ontology may be scientifically sophisticated, legally recognised, administratively efficient, culturally dominant, or supported by consensus. None of these properties substitutes for continued answerability to evidence, counterevidence, level discipline, affected-party knowledge, and correction.

In ADM–CIV terms, epistemological priority prevents the administrative-managerial position from treating its current reality-picture as the final boundary of legitimate civil evidence. It does not abolish ADM closure. It requires that closure remain temporary, reason-giving, reviewable, and reversible.

11.4.2 Normative genealogy: objective-critical reason

The normative genealogy of this principle lies in Horkheimer’s distinction between objective and subjective reason and in Horkheimer and Adorno’s broader critique of Enlightenment rationality (Horkheimer 1947; Horkheimer and Adorno 2002).

Subjective reason evaluates the suitability of means relative to ends already given to the subject, organisation, market, state, or technical system. Instrumental reason is its operationally narrowed form: calculation, adaptation, prediction, administration, efficiency, and control. These capacities are necessary. The failure begins where they exhaust the meaning of reason.

Objective reason raises a different question: whether the end itself possesses rational standing. It permits reason to judge the purposes of the subject rather than serving them automatically. Dialectic of Enlightenment supplies the complementary historical diagnosis: rationalisation may expand while the capacity of reason to criticise its own purposes contracts, allowing calculation, classification, and administration to become vehicles of domination.

This article does not attribute the phrase “priority of epistemology over ontology” to Horkheimer or Adorno. Their critique provides the normative source from which the principle is reconstructed as a governance requirement.

11.4.3 Exact relation: normative source and corrective constitution

Objective-critical reason and epistemological priority answer different questions.

Objective-critical reason asks:

What is reason authorised to judge?

Its answer is that ends, institutions, collective identities, interests, and systems remain within the jurisdiction of reason.

Epistemological priority asks:

How must a decision regime be constituted so that this judgment can remain consequential?

Its answer is that no ontology, objective, model, or institutional reality-picture may define in advance the admissible evidence by which it can be challenged.

The relation may be represented as:

Objective-Critical Reason→normative translationEpistemological Priority→institutional embodimentCorrection Sovereignty→operative implementationRevisable Permission.

(29)

Objective reason is therefore the normative source. Epistemological priority is the corrective constitution. Correction Sovereignty is the authority form through which that constitution becomes decision-bearing. Revisable permission is its operative result.

This translation also marks the article’s extension beyond the Frankfurt diagnosis. The critique of ends is converted into an institutional architecture capable of asking which evidence may reopen the ontology, who possesses standing, which authority may change permission, and how reversal and restoration are implemented.

11.4.4 Functional exclusivity and necessary insufficiency

The principle is exclusive only in a functional sense.

Technical competence cannot substitute for it. Professional expertise cannot substitute for it. Consensus cannot substitute for it. Transparency, consultation, audit, or model evaluation cannot substitute for it where accepted ontology determines in advance what those procedures may question.

Accordingly:

GR⇒EP,EP⇏GR,

(30)

where GR denotes governance reliability and EP epistemological priority.

The first relation states the non-substitutable claim: a governance regime cannot be reliably corrective if its governing ontology is immunised from epistemological reopening.

The second states the insufficiency claim: epistemological priority may remain symbolic unless it is joined to standing, evidence access, Correction Sovereignty, implementation ownership, reversibility, restoration, and institutional learning.

The article therefore does not propose epistemological priority as a self-executing philosophical ideal. It proposes it as the indispensable upstream condition of a complete correction architecture.

11.4.5 Contemporary stress test: collective projection and scientific legitimation

Two contemporary questions test the principle without becoming load-bearing premises of the article.

The first asks whether modern public ontology reproduces a collective ontogenesis projection: the transfer of organismic-developmental grammar into the interpretation of species, genes, civilisation, and the universe. The hypothesis concerns cross-scale explanatory form, not literal belief that a species, gene, civilisation, or universe is an organism. It remains an open research question. OPI is retained only as a diagnostic candidate for identifying possible loss of object, mechanism, level, or metaphor boundary; it does not establish a biological cognitive type or collective causal mechanism (Dunavich 2026h, 2026i).

The second asks whether some institutional uses of scientific theory create a scientific-legitimation interface. The relevant possibility is not that a theory is fabricated as political packaging. It is that the legitimate authority of a bounded theory may be transferred into a broader ontology and then into permission without independent justification at each bridge.

The candidate sequence is:

Bounded Scientific Claim→Authority Transfer→Expanded Public Ontology→Institutional Permission.

This sequence is diagnostic, not accusatory. It is weakened or rejected where the wider ontology follows from explicit bridging evidence, where limitations remain visible, where rival models retain standing, and where permission is justified independently of scientific prestige.

These stress tests clarify why epistemological priority must govern not only whether a claim is scientifically legitimate within its domain, but also whether the authority of that claim is entitled to travel across explanatory levels and into the ADM–CIV permission structure.

11.5 Correction hierarchy

The general correction hierarchy is:

output correction;
model correction;
evaluation correction;
representation correction;
ontology correction;
authority correction;
governance correction.

A lower-level correction may be adequate. Not every output error requires ontology revision. The governing rule is:

Correction must reach the highest level at which the defect is generated.

An output-level fix cannot resolve an ontology-level defect. A new explanation cannot resolve an authority-level defect. An appeal cannot resolve a policy-level defect where the appellate body lacks power to change the rule.

Once ontology is translated into permission, the central question becomes institutional:

Who possesses the authority to translate ontology into permission, and who possesses the authority to reopen it when its consequences become defective?

Part IV — Institutional Reproduction

12. ADM, CIV, and the Three Sovereignties

Institutional decision regimes contain two recurrent functional positions. They are not permanent social classes, moral identities, or fixed organisations.

12.1 ADM and CIV

ADM is the administrative-managerial position. It translates evidence, policy, risk, organisational purpose, resource constraint, and legal authority into an operative state. ADM is responsible for continuity, coordination, implementation, protection, recordkeeping, and decision closure.

CIV is the civil, exposed, dependent, or cost-bearing position. It encounters the integrated consequence of the decision regime and may contribute affected-party knowledge, external evidence, criticism, alternative object definitions, implementation-failure signals, and demands for reconsideration.

The same actor may occupy both positions in different episodes. A clinician may be ADM toward a patient and CIV toward hospital administration. A software engineer may implement a policy while being exposed to an executive release decision. The distinction concerns function in a specified episode.

ADM without CIV→Correctionless Filtering,

CIV without ADM→Criticism without Implementation.

(31)

The governance objective is not victory of one side over the other. It is a complete correction path connecting decision, consequence, evidence, review, authority, and implementation.

12.2 Forward and reverse chains

The forward chain is:

ADM Classification→Permission→Implementation→CIV Consequence.

(32)

The reverse chain should be:

CIV Consequence→Evidence→Substantive Review→Correction Authority→ADM State Change.

(33)

Failure frequently occurs in the reverse chain.

12.3 Decision Sovereignty

Decision Sovereignty is authority to issue the initial operative determination. It includes control over problem definition, data, criteria, threshold, and initial permission. It may be distributed across designers, managers, regulators, evaluators, clinicians, procurement bodies, and software systems.

12.4 Correction Sovereignty

Correction Sovereignty is authority and capacity to reopen the object, reconsider evidence, alter the threshold, reverse the decision, require implementation of reversal, restore the affected position, and revise the generating rule. Formal appellate authority is insufficient where implementation remains optional or diffuse.

12.5 Public-Grammar Sovereignty

Public-Grammar Sovereignty is authority to define the categories through which criticism, harm, safety, dignity, identity, evidence, and legitimacy become institutionally intelligible. An institution may permit appeal while defining the vocabulary inside which appeal must occur.

The highest-risk configuration is:

DSADM↑+PGSADM↑+CSCIV↓.

(34)

Actors occupying the ADM position in the specified episode control the decision, the grammar through which it is understood, and the evidentiary threshold for correction, while actors occupying the CIV position bear consequence without an operative correction path.

12.6 Concentration and fragmentation

Concentration is not the only failure. Correction Sovereignty may be fragmented so completely that no actor owns the full reversal. One body reviews evidence. Another changes the record. A third restores access. A fourth controls policy. Every actor may perform its assigned role while correction remains incomplete.

This produces orphaned correction.

12.7 Protected Contestability

Protected Contestability is the institutional path through which criticism can be raised, preserved, heard, tested, translated into an authorised decision, implemented, and restored without requiring the critic or affected party to absorb retaliation, exclusion, or impossible evidential burdens.

It requires standing, evidence preservation, a competent forum, a visible permission owner, appeal, implementation, and restoration. The sequence is non-equivalent:

Expression≠Standing≠Review≠Correction≠Restoration.

(35)

Accountability and contestability literature provides important antecedents concerning forums, judgment, consequence, responsibility fragmentation, multiple-accountabilities disorder, and rights to contest consequential automated decisions (Bovens 2007; Koppell 2005; Abbott and Snidal 2009; Kaminski and Urban 2021). The Three Sovereignties and Protected Contestability remain article-specific constructs.

13. Affective Jurisdiction and Criticism-Governance

Affective jurisdiction is used here as a bounded institutional stress test rather than as a universal theory of criticism. It makes the distinction between evidence, authority, object definition, and permission unusually visible. The local mechanism developed in this section is one instance of the wider governance problem: a valid signal acquires authority beyond the question it can answer and is then translated into an operative permission effect.

Affective evidence can be indispensable where the governed object includes lived harm, humiliation, fear, dependency, exclusion, or experienced consequence that cannot be reconstructed adequately from administrative records alone. Yet affective evidence has a jurisdiction.

The fact that an experience is authentic does not determine every proposition associated with it:

Serious Affective Uptake≠Affective Sovereignty over Criticism.

(36)

First-person experience may possess strong authority concerning felt humiliation, fear, pain, exclusion, or dependency. It may possess partial authority concerning social meaning, likely behavioural consequence, or group experience. It does not independently settle external causal structure, truth of the criticised claim, moral status of the critic, or public permission appropriate to the criticism.

13.1 Affective-Jurisdiction Failure

Affective-Jurisdiction Failure occurs where authority attached to an affective report expands beyond the question the report can answer. The expansion may move from experience to causation, causation to object judgment, object judgment to moral classification, or moral classification to institutional permission.

The failure is not that the report is emotional. The failure is jurisdictional transfer.

13.2 Sentimental Veto classifier

For a specified episode e, a strict Sentimental Veto classification requires four jointly supported conditions:

SVe⇔ARe∧AJe∧MDDe∧PEe,

(37)

where AR is an identifiable affective report; AJ affective authority exceeding its proper jurisdiction; MDD displacement of the material or substantive decision domain; and PE a consequential Permission Effect.

The classifier is conjunctive. An affective report without jurisdictional expansion is not a Sentimental Veto. A jurisdictional error without operative consequence is not a complete episode. A protective intervention that preserves substantive criticism may be legitimate protection rather than insulation.

13.3 Object inflation and permission effects

A bounded object may expand through institutional processing. A report concerning one experience becomes a judgment concerning an entire topic, class of claims, person’s legitimacy, field’s permission, or group’s access. This is object inflation.

The permission-effect profile is:

PE=(O,I,S,D,R,Med),

(38)

where the effect may concern object access, institutional inclusion, speech or publication, decision participation, role or resource, and medium or representational status.

A veto need not take the form of explicit censorship. It may operate through relocation, delay, loss of salience, exceptional review, altered classification, reputational marking, or removal from the authoritative medium.

13.4 Protection and permission registers

The institution should maintain two distinct registers:

Protection register: what immediate protective response is justified?
Permission register: what may happen to the substantive object, criticism, speaker, workflow, or policy?

A temporary protective hold may be appropriate while the substantive claim remains available for review. The error occurs when protection silently determines permission.

Critical Tolerance therefore requires serious uptake of affected experience, preservation of the criticised object, question-specific evidentiary authority, independent review of permission effects, symmetric access to correction, and restoration and policy learning where error is found.

Its governing principle is:

No credibility deficit and no testimonial sovereignty.

The Sentimental Veto classifier is not the general mechanism of institutional closure developed in this article. It is a local classifier whose recurrence may contribute to a pattern only where common workflow, authority practice, or public grammar can be demonstrated.

14. Durable Correction Asymmetry

An institutionalised pattern becomes a closure regime only where correction is durably more difficult than creating or preserving the operative effect.

Durable Correction Asymmetry, or DCA, is the persistent structural difference between the path that creates, maintains, or renews a permission state and the path capable of reopening, reversing, implementing reversal, restoring the affected position, and revising the generating rule.

DCA is not merely delay, bureaucracy, or an unsuccessful appeal. It is analysed across eight dimensions:

Standing asymmetry: evidence supporting restriction can initiate action while critics or affected decision subjects cannot initiate substantive reconsideration.
Evidence asymmetry: evidence supporting the operative state enters the authorised record while contrary, contextual, or affected-party evidence lacks an equivalent route.
Threshold asymmetry: a low or precautionary threshold creates a durable effect while reversal requires near-conclusive proof.
Authority asymmetry: many actors can preserve the decision but no actor owns operative reversal.
Implementation asymmetry: restriction is immediate and automatic while restoration is optional, delayed, or distributed.
Temporal asymmetry: a provisional measure persists beyond its justification, review date, or stated expiry.
Policy asymmetry: an individual outcome can be appealed but the generating rule cannot be reviewed.
Representational asymmetry: the initiating side defines the event, category, relevant harm, and public narrative, while the correcting side must first defeat that representation before reaching the decision itself.

Three stages must remain distinct:

Reversal≠Implementation≠Restoration.

(39)

Let P(S,T) denote an institutionalised pattern for system S during period T, and DCA(S,T) durable correction asymmetry. A closure regime requires:

CRS,T⇔PS,T∧DCAS,T.

(40)

The expression is schematic. It does not establish prevalence, causal magnitude, or equilibrium.

Correction paths need not be perfectly symmetric. Asymmetry may be justified by emergency, confidentiality, precaution, irreversibility, or unequal false-positive and false-negative costs. The normative question is whether the asymmetry remains proportionate, reviewable, time-bounded, and connected to restoration and policy learning.

15. Corrective-Voice Attrition

Institutions are shaped not only by the decisions they make but by which participants remain willing and able to challenge those decisions. Organisational research shows that participants may withhold problem-relevant information where speaking appears risky, futile, disloyal, or inappropriate. Psychological safety and perceived managerial openness affect voice and learning behaviour (Hirschman 1970; Edmondson 1999; Morrison and Milliken 2000; Detert and Burris 2007; Detert and Edmondson 2011; Morrison 2014).

The article proposes Corrective-Voice Attrition.

15.1 Definition

Corrective-Voice Attrition occurs where participants capable of sustained, error-relevant challenge move disproportionately toward silence, accommodation, reduced participation, reassignment, or exit because voice lacks a credible route to operative correction.

The participants need not be morally superior or substantively correct. The loss is functional: fewer actors remain available to preserve adversarial evidence and test the institution’s governing frame.

15.2 Candidate mechanism

Error-Relevant Observation→Attempted Voice→Symbolic or Non-Operative Review→Anticipated Futility or Cost→Reduced Future Voice→Selective Loss of Corrective Participants→Weaker Evidence Environment→Apparent Consensus.

(41)

The article proposes that repeated non-operative voice may contribute to selective loss of correction-relevant participation. Whether the mechanism is sufficient, necessary, or independent of ordinary attrition remains an empirical question.

15.3 Denominator discipline

The denominator is not all employees, users, or citizens. The relevant population includes participants who possess correction-relevant evidence or expertise, have standing or practical access to raise it, and could contribute to correction if the path were credible.

A defensible study should distinguish eligible corrective participants, participants attempting voice, participants receiving substantive review, participants obtaining operative correction, participants reducing participation or exiting, and participants remaining while withholding future voice. Without this denominator, ordinary attrition may be misclassified as selective loss of correction capacity.

15.4 Silence, voice, and consensus

Silence⇏Agreement,

Voice⇏Influence⇏Correction.

(42)

Observed agreement may contain authentic endorsement, strategic accommodation, silence, role conformity, and selection through exit. Quiet cannot be interpreted as consensus without independent evidence concerning participation, channel credibility, exit, and correction.

15.5 Adversarial memory

Institutions require adversarial memory: retained knowledge of prior objections, alternative object definitions, failed policies, suppressed anomalies, and unrealised correction paths. Corrective-Voice Attrition weakens that memory. A later generation may encounter the existing state as settled consensus without access to the objections removed during its formation.

15.6 Observable implications and falsifiers

Candidate indicators include declining repeated use of complaint channels, concentration of voice among newcomers, departure after unsuccessful correction attempts, reduced evidentiary diversity in later decisions, and widening gaps between private concern and public record.

The construct is weakened where departures track compensation or ordinary mobility, participants who remain possess equal or greater corrective expertise, channels retain demonstrated credibility, correction rates remain stable, or independent evidence shows no reduction in error-relevant information.

16. CEP-Consistent Persistence

Section 8 distinguished repeated outcome, institutionalised pattern, closure regime, and CEP-consistent persistence. This section addresses only the final transition. CEP becomes admissible after a closure regime has been established independently; it is not a new name for recurrence, path dependence, or institutional inertia.

16.1 Stage One — Closure and rival explanation

The analysis must first establish a recurrent reproduction rule, durable correction asymmetry, and failure of simpler rival explanations. Relevant rivals include path dependence, hierarchy, legal constraint, resource scarcity, principal–agent problems, audit incentives, organisational silence, risk aversion, and legitimate precaution.

If these explain persistence adequately, CEP should not be invoked.

16.2 Stage Two — Strategic persistence

Only after closure has been established may the study examine actor-specific benefits of continuation, expected responses of other actors, costs of unilateral deviation, shared expectations that correction attempts will fail or be punished, and persistence through genuine, feasible correction opportunities.

CEPS,T=(CR,LIC,DC,SE,PC),

(43)

where CR is a closure regime independently established; LIC continuation that is locally incentive-compatible; DC meaningful deviation cost; SE sufficiently shared expectations; and PC persistence across genuine correction opportunities.

An actor need not prefer the complete institutional outcome. Continuation may remain locally rational because the actor preserves role security, avoids unilateral cost, maintains professional legitimacy, protects coordination, or expects others not to change.

A regime regarded by relevant participants as collectively inferior, inefficient, or burdensome may nevertheless persist because no actor can alter it unilaterally at acceptable local cost. A Pareto claim requires an independently specified feasible alternative and actor-level preference evidence.

16.3 Critique absorption, selection, and apparent consensus

Critique absorption may stabilise the regime by lowering external pressure, preserving the appearance of openness, and allowing actors to continue without confronting the governing ontology. Corrective-Voice Attrition can further stabilise the system as the evidence environment narrows and apparent consensus increases.

The full recursive relation is:

Public Ontology→Public Grammar→ADM Classification→Permission Effect→CIV Burden→Corrective Voice→Correction Asymmetry→Voice Attrition→Apparent Consensus→Reinforced Public Ontology.

(44)

16.4 Ontology defines the game

Ontology defines the players, strategies, recognised payoffs, legitimate evidence, and available deviations. The repeated game then generates institutional investment, public expectation, data, and professional roles that make the ontology appear increasingly natural.

Ontology Defines Game⇄Game Reproduces Ontology.

(45)

A CEP-consistent diagnosis does not imply conspiracy, bad faith, shared ideology, or deliberate coordination. Actors may behave transparently and locally rationally. The analytical claim is structural: the decision regime rewards continuation more reliably than justified correction, and relevant actors anticipate that unilateral deviation will not succeed.

CEP should be narrowed, simplified, or rejected where path dependence explains the evidence, actor incentives cannot be identified, deviation costs are absent, expectations are not shared, or genuine correction opportunities did not occur.

Part V — Comparative Test and Model Revision

17. A Multidimensional Comparative Test of Ten High-Capacity Profiles

The comparative purpose of this section is not to determine which thinkers possess formal reason and which do not. Every selected profile demonstrates advanced formal capacity within at least one publicly documented intellectual domain.

The test instead asks:

When formal capacity is held approximately constant at a high level, does it appear in one general form, two stable directional types, or multiple configurations of utilisation?

The inquiry concerns public textual-intellectual profiles rather than complete persons. It examines published work, public lectures, methodological statements, institutional proposals, and documented intellectual roles. It does not infer private motivation, neurological structure, moral character, or total cognitive identity.

17.1 Initial two-cluster hypothesis

The inherited research design distinguished two candidate clusters.

The first emphasised boundary discipline, mechanism, conceptual restraint, resistance to explanatory inflation, and protection of justification conditions. The second emphasised narrative decomposition, religion critique, historical consciousness, public memory, symbolic authority, and the social production of collective meaning.

The distinction did not rank intelligence. Both candidate clusters were selected from profiles of demonstrably high formal capacity.

The initial hypothesis attached one characteristic risk to each cluster. The boundary-oriented cluster was expected to risk non-conversion: the human, political, or institutional domain might be recognised without being translated into an operative correction programme. The social-symbolic cluster was expected to risk ontogenesis projection: historical sequence might be converted into maturation, cultural transformation into organismic development, or collective memory into destiny.

These predictions were analytically coherent and falsifiable. The primary-corpus test did not support them as the final classification of the ten purposively selected profiles. The predicted separation and characteristic risk distribution did not appear in the coded corpus. This result rejects the binary model for the present article; it does not establish that no comparable clustering could be found in a different population or design.

17.2 Case-admission protocol

A profile entered the comparison only where the available corpus satisfied enough of the following conditions:

Capacity anchor — advanced formal competence was independently established.
Primary corpus — direct public work was available rather than commentary alone.
Cross-domain contrast — at least two objects, genres, or intellectual roles could be compared.
Direct access — substantive passages or reliable full-text material were available.
Counterevidence — the corpus permitted active search for material inconsistent with the inherited classification.
Language control — translation, terminology, and edition differences could be handled without attributing unsupported claims.

The cases did not satisfy these conditions equally. Unequal access is part of the evidentiary result. A profile cannot receive a stronger classification because it fits the theory more elegantly.

17.3 Final coding dimensions

The binary classification is replaced by a multidimensional profile:

Pi=(FC,B,M,SP,DI,SSR,MR,CS,IC),

(46)

where FC is the formal-capacity anchor; B boundary discipline; M mechanism specification; SP source–path discipline; DI developmental integration; SSR social-symbolic reach; MR medium and representational discipline; CS corrective symmetry; and IC institutional conversion.

Formal capacity functions primarily as the admission threshold. The remaining dimensions describe the direction and conversion of that capacity.

Boundary discipline concerns distinctions among concept, metaphor, model, ontology, evidence, authority, local claim, universal conclusion, object, and representation.

Mechanism specification concerns operative relations, constraints, processes, and transitions rather than temporal sequence or retrospective coherence alone.

Source–path discipline concerns distinctions among source, generative capacity, realised path, and contingency.

Developmental integration concerns formation, historical sequence, transition, emergence, and continuity without converting change into intrinsic maturation.

Social-symbolic reach concerns identity, religion, legitimacy, memory, obligation, narrative, and institutional meaning.

Medium and representational discipline concerns how image, narrative, interface, classification, and public form change the operative meaning of a claim.

Corrective symmetry concerns whether constitutive assumptions receive scrutiny comparable to that directed at external objects.

Institutional conversion concerns whether analysis becomes an operative correction programme. Conversion may be experimental, educational, cultural, administrative, legal, or political.

17.4 Comparative matrix

Table 1. Multidimensional Comparative Matrix of Ten High-Capacity Profiles

The classifications below are qualitative research judgments rather than psychometric measurements.

Legend: VS = very strong; S = strong; MS = moderate–strong; M = moderate; I = indeterminate or inadequately tested; NA = not a legitimate inference from the available corpus.

Profile	B	M	SP	DI	SSR	MR	CS	IC	Evidence confidence
Robert J. Aumann	S	S	S/MS	M	MS	M	I	MS	Medium–high
Yeshayahu Leibowitz	S	S	S	M	S	M	MS	S	High
Jacob D. Bekenstein	S	S	S	M	NA	M	S within science	NA	High, domain-bounded
Yuri P. Altukhov	S	S	S	M	NA	NA	I/M	M	Medium
Elisha Haas	VS	VS	S	S	NA	M	MS within science	S experimentally	High, domain-bounded
Michael Harsegor	MS	S	MS	S	S	S	I/M	M	Medium–low
Henry Unger	VS	S	MS	MS	S	VS	M	M	Medium
Moshe Zuckermann	VS	VS	S	MS	VS	S	MS	M	Medium–high
Avishai Ehrlich	MS	S	MS	S	VS	S	MS	M	High
Yigal Bin-Nun	S	MS	S	VS	VS	S	M	M	Medium–high

No five-versus-five separation appears:

SSR↑⇏B↓,DI↑⇏SP↓,B↑⇏IC↓.

(47)

17.5 Aumann and Leibowitz: the two controls

Robert J. Aumann and Yeshayahu Leibowitz originally entered the study as controls. Aumann prevents uneven utilisation from being reduced to weak intelligence. Leibowitz prevents cross-domain separation from being reduced to ignorance. Those control functions survive. The stronger cluster inferences do not.

Robert J. Aumann

Aumann’s technical corpus establishes exceptional capacity in formal abstraction, strategic reasoning, repeated games, common knowledge, incomplete information, and equilibrium analysis. His work applies game-theoretic reasoning to war, deterrence, cooperation, loyalty, punishment, enforcement, and long-run strategic stability (Aumann 1976, 2005; Nobel Prize Outreach 2005).

The corpus therefore does not support a broad claim that his formal capacity remains confined to technical domains. The reviewed public corpus leaves the degree of cross-domain corrective symmetry unresolved. Aumann is classified as a mixed maximal-capacity control with strong technical and strategic conversion but indeterminate corrective symmetry across religious and political commitments.

Yeshayahu Leibowitz

Leibowitz supplies the strongest counterevidence to the inherited non-conversion hypothesis. His distinctions among science and value, religion and state, historical event and redemption, governmental authority and religious authority, and symbolic declaration and practical implementation are repeatedly translated into institutional criticism (Leibowitz 1992).

His programme includes limitation of state authority, separation of religion from government administration, independent religious institutions, civil legal arrangements, criticism of official religious bureaucracy, and technical reorganisation of Sabbath observance in a modern economy.

The reasoning chain is:

Boundary Distinction→Authority Analysis→Burden Reconstruction→Institutional Redesign.

(48)

Leibowitz is therefore not merely a non-ignorance control. He is a direct case of boundary discipline becoming institutional correction.

17.6 Bekenstein and Ehrlich: ontology and historical structure

Jacob D. Bekenstein

Bekenstein strongly exemplifies model-boundary discipline, mechanism specification, source-level uncertainty, and adversarial scientific correction. Yet his work is not ontologically timid. Black-hole thermodynamics is extended toward entropy bounds, information theory, holographic description, and possible fundamental physical ontology (Bekenstein 1973, 1981, 2003).

The relevant feature is not absence of extension. It is disciplined extension: established relation is distinguished from proposed interpretation; general law from implementing mechanism; known domain from speculative extension; and current fundamental model from final ontology.

His profile is best described as disciplined ontological extension rather than boundary restraint without development. His later work also tests governing principles through adversarial thought experiments, supporting strong scientific corrective symmetry. The corpus does not justify inference concerning civil institutional conversion because that is not the object of the relevant works.

Avishai Ehrlich

Ehrlich strongly exemplifies social-symbolic and historical integration. His analysis includes religion, identity, nationalism, collective memory, state formation, and political legitimacy while also specifying mechanisms involving colonial borders, oil regimes, military infrastructure, international finance, superpower intervention, coalition dependence, and recursive violence (Ehrlich 1992, 2003).

Historical stages are not presented as organismic maturation. They are connected to changes in sovereignty, institutions, military capacity, economic relations, and external patrons. Ehrlich therefore provides counterevidence to the presumed opposition between narrative interpretation and causal mechanism.

17.7 Altukhov and Bin-Nun: source discipline across different objects

Yuri P. Altukhov

Altukhov’s population genetics makes stability an independent explanatory object. His corpus asks not only how gene frequencies change but under which conditions differentiated population systems preserve adapted genetic organisation across generations (Altukhov 2006).

The work combines variation, selection, migration, polymorphism, population hierarchy, and stability. It does not support static essentialism. Nor does it support treating population-genetic change as a complete explanation of developmental or species-level architecture.

The strongest source-supported conclusion is that Altukhov distinguishes population variability from the conditions maintaining structured and reproducible population systems. The more extensive architecture–frequency implications developed by this article remain synthetic rather than direct Altukhov claims.

Yigal Bin-Nun

Bin-Nun’s work combines religion critique with extensive source criticism. He distinguishes narrated event from time of composition, biblical Israelite from later Jewish identity, ancient cult from later religion, theological periodisation from imperial-political periodisation, and canonical unity from layered textual production (Bin-Nun 2016, 2023).

His developmental histories are tied to textual composition, imperial transition, institutional destruction, political reform, and interaction among religious traditions. This is developmental integration. It is not sufficient evidence of ontogenesis projection.

The central open question concerns modal calibration. A scholar may apply strong scepticism to inherited narratives yet state replacement reconstructions more categorically than sparse evidence warrants. Bin-Nun therefore provides strong source discipline, strong category revision, and an open test of corrective symmetry between rejected and replacement ontologies.

17.8 Haas and Unger: structuralism across mechanism and representation

Elisha Haas

Haas studies how amino-acid sequence becomes three-dimensional protein structure. The loop hypothesis proposes that early non-local interactions close long loops, reduce conformational possibility, and constrain subsequent folding pathways (Orevi et al. 2013; Bergasa-Caceres, Haas, and Rabitz 2019).

The corpus distinguishes final structure, thermodynamic constraint, kinetic route, local interaction, non-local interaction, and transient intermediate. The explanatory chain is:

Sequence→Early Constraint Architecture→Folding Path→Native Structure.

(49)

The major strength is experimental conversion. Hypotheses concerning early interactions are translated into site-specific labelling, time-resolved measurement, mutational intervention, and pathway comparison.

The phrase “second genetic code” must nevertheless remain bounded. The corpus supports research-program language concerning the sequence-to-structure relation. It does not establish a second discrete code possessing a fully specified syntax and translation rule.

Henry Unger

Unger studies artistic and cinematic representation through logic, medium, rhetoric, composition, perception, and aesthetic judgment. His analysis of Escher distinguishes the coherent pictorial surface from the globally impossible world represented through it (Unger 1999; Tel Aviv University 2017).

Locally Coherent Presentation+Globally Inconsistent Representation=Pictorial Nonsense.

(50)

This is a formal model/object distinction. Unger’s corpus also pursues rigorous criteria of aesthetic judgment and therefore contradicts classification as primarily relativistic, non-mechanistic, or structurally loose.

Haas and Unger share a formal problem: components do not explain organisation automatically. For Haas, amino-acid sequence does not transparently reveal folding path. For Unger, individual pictorial elements do not establish the coherence of the represented world.

17.9 Harsegor and Zuckermann: narrative, memory, and institutional mechanism

Michael Harsegor

Harsegor’s public method is strongly narrative. His academic objects are also institutional. They include royal councils, elite concentration, oligarchic circulation, historical periodisation, and decision under crisis (Harsegor 1994; Tel Aviv University n.d.).

His work on oligarchy identifies recurrent relations among access to power, elite formation, democratic mobilisation, and renewed concentration. His crisis-decision corpus directs attention toward actor knowledge, constraint, judgment, contingency, and counterfactual possibility.

The profile therefore appears to combine narrative integration, institutional mechanism, and contingent agency. Because substantial portions of his academic and broadcast corpus were unavailable for passage-level analysis, the classification remains provisional and access-limited.

Moshe Zuckermann

Zuckermann’s work combines collective memory, Holocaust reception, identity, political culture, religion, art, and ideology with strong institutional and representational distinctions. He separates historical event, historiography, public memory, political deployment, and legitimacy function (Zuckermann 1998, 2002, 2006).

One historical event may acquire different public functions within different political cultures:

One Historical Event+Different Institutional Regimes→Different Public Meaning Functions.

(51)

This supports the article’s distinction between truth-value and public deployment.

The reviewed corpus does not establish whether Zuckermann’s governing critical-theory framework is subjected to equivalent adversarial testing. This is an open comparative question, not a finding of corrective asymmetry.

17.10 Comparative result

The inherited two-cluster model fails as the final result of this article in three ways.

First, boundary, mechanism, development, representation, and social-symbolic reach coexist across the supposed divide.

Second, the characteristic risks do not distribute as predicted. Non-conversion is contradicted most strongly by Leibowitz. Ontogenesis projection is not established in the five social-symbolic profiles.

Third, discipline is not equivalent to formal direction. Mechanism appears in physics, biology, aesthetics, history, and political sociology. Developmental integration appears in molecular folding, population structure, religious history, political history, and cultural memory.

The initial model was not analytically empty. It identified two genuine intellectual functions: protection of justification boundaries and reconstruction of social-symbolic intelligibility. The error was to treat the functions as opposing profile families. The two-cluster model remains only as the historical heuristic through which the stronger multidimensional model was discovered.

The bounded conclusion is:

High formal capacity may be realised through different configurations of boundary control, mechanism, source discipline, developmental integration, social-symbolic reach, representational discipline, corrective symmetry, and institutional conversion. No one configuration guarantees governance reliability.

Governance relevance may be represented schematically as:

GR=f(B,M,SP,DI,SSR,MR,CS,IC,Auth,Rev),

(52)

where Auth denotes operative authority and Rev reversibility and restoration.

18. Alternative Explanations, Access Limits, and Defeat Conditions

The ten-profile analysis does not establish a population theory. The cases are purposive rather than statistically representative. Their function is to test whether an inherited conceptual distinction survives examination in difficult cases where formal capacity cannot plausibly be treated as absent.

18.1 Rival explanations

Discipline may explain differences because physicists, molecular biologists, historians, philosophers, and political sociologists work with different objects, evidence, causal scales, and professional standards.

Genre may change citation density, uncertainty language, narrative compression, polemical force, and explicit treatment of rivals.

Audience may require simplification, which is not necessarily distortion. The relevant question is whether assumptions and uncertainty remain visible.

Historical context may alter reasoning under war, institutional crisis, disciplinary controversy, religious conflict, or a specific political regime.

Political or religious commitment may influence problem selection, evidence salience, burden attribution, and interpretation of institutional purpose without determining one intellectual profile.

Institutional role affects the highest available level of conversion. A theoretical physicist is not expected to provide a civil appeals architecture. A regulator or deployer faces different implementation criteria.

Professional incentives may shape which questions remain safe and whether constitutive assumptions receive scrutiny equivalent to that applied to external objects.

Rhetoric and translation may make research-program language appear stronger or weaker than the underlying argument.

Selection bias arises because the cases were selected for conceptual richness and apparent relevance to the inherited distinction.

Confirmation bias arises because a rich vocabulary can absorb ambiguous evidence. Cluster labels must be outputs of analysis, not inputs controlling interpretation.

18.2 Within-person heterogeneity

A person-level summary may conceal major internal variation. A profile should therefore be treated as a distribution across domain, genre, period, and institutional role:

Pi={Pi,d,g,t,r}.

(53)

Where within-profile variation is large, a single profile label may be misleading.

18.3 Public profile and private cognition

The corpus concerns public work. Public reasoning may reflect editorial constraint, strategic silence, legal risk, institutional role, or audience design. No private cognitive diagnosis follows.

18.4 Causal underdetermination

The comparison establishes variation. It does not establish its source. Possible causes include training, biography, ideology, institution, socialisation, medium, authority, and prior explanatory orientation. DIC remains one speculative possibility. It is not supported by the comparative distribution itself.

18.5 Defeat conditions

The multidimensional model should be narrowed, simplified, or rejected if:

independent coders cannot distinguish the dimensions reliably;
coding scores merely reproduce disciplinary categories;
genre and role explain all observed differences;
the same corpus receives unstable coding across rounds;
profile summaries conceal more variation than they explain;
the dimensions fail to predict any difference in correction behaviour;
access-complete corpora reverse the principal findings;
a simpler model explains the evidence equally well.

DIC should be removed from the research programme if it cannot predict variance beyond discipline, genre, role, medium, institution, socialisation, ideology, and incentive, or if the proposed orientations lack developmental stability, within-person consistency, or predictive value.

The ten cases permit one clear result: the simple binary model is not supported as the final classification of this corpus. They permit one provisional replacement: advanced formal reason differs along multiple independently varying dimensions. They do not establish a validated measurement instrument, population distribution, causal origin, or general theory of cognitive development.

Part VI — Existential Extensions

19. Intersubjective Intensification under Uncertainty

Uncertainty does not merely reduce information. It changes the value of coordination. When individuals and institutions cannot predict future conditions, others’ behaviour, resource availability, institutional protection, or political stability, shared reality, identity, norm, and authority may become more valuable.

This does not mean that uncertainty automatically produces conformity, nationalism, religion, or authoritarianism. The response depends on prior institutions, available identities, threat representation, leadership, material conditions, and correction architecture.

The article proposes the following candidate conditional sequence:

Uncertainty→Higher Value of Shared Orientation→Changed Utilisation Incentives→Higher Cost of Constitutive Dissent.

(54)

Under uncertainty, formal capacity may remain intact while its use changes. Reason may be redirected toward defending coordination, rationalising authority, protecting group identity, interpreting ambiguity as threat, or optimising action inside a narrowed frame.

FCt≈FCt+1whileDUt≠DUt+1.

(55)

Research on uncertainty and group identification suggests that uncertainty may increase attraction to clearly defined groups and prototypical norms under specified conditions (Hogg 2007). Cross-cultural work also suggests that threat can be associated with norm tightening, though the direction and magnitude vary across societies and institutions (Gelfand et al. 2011).

These literatures support only component relations. They do not validate the complete sequence proposed here.

Strong identity, religion, or national solidarity does not establish a closure regime. Closure requires recurrent reproduction, durable correction asymmetry, and—at the strongest level—strategic persistence. A community may possess intense shared content while maintaining protected criticism, open evidence, plural authority, and reversible policy.

Material dependence may change the cost of dissent. Employees, patients, citizens, users, or contractors may possess advanced reasoning yet refrain from constitutive criticism because exit is costly, authority controls essential resources, or correction channels lack credibility. This is an incentive and governance claim, not a claim that dependency suppresses abstract cognition.

The governance requirement under uncertainty is not permanent openness. It is provisional closure with preserved correction: a decision may be temporarily fixed while the ontology remains reviewable, revisable, and explicitly bounded by uncertainty.

The two-layer corrective principle explains why this arrangement is not indecision. Objective-critical reason preserves jurisdiction over the end; epistemological priority preserves the revisability of the ontology through which the end becomes policy. ADM may therefore close a decision for action without converting temporary coordination into permanent truth. Where CIV evidence retains standing and a route to operative correction, stability can be produced with less dependence on suppressive closure.

20. Ontological Lock-In and the Conditional Path to War

War is multicausal. No direct path runs from developmental cognition, PIC, religion, identity, sentiment, or ontological closure to armed conflict. War may arise through security dilemmas, territorial disputes, resource competition, elite strategy, alliance structures, miscalculation, state weakness, historical grievance, institutional collapse, or deliberate aggression.

The article proposes a narrower conditional mechanism concerning the process through which threat, identity, public ontology, and correction-path contraction may make escalation more difficult to reverse.

20.1 Threat is interpreted

A threat is not only an external event. It is also a classified event. The institution determines what counts as threat, who represents it, how imminent it is, which population is implicated, and which responses are legitimate.

This does not imply that threat is merely constructed. Material threat may be severe and independently verifiable.

Material Threat≠Threat Representation,

Governance Response=f(Material Evidence,Representation,Authority,Institutional Interest).

(56)

20.2 Identity–claim fusion and sacred value

Under intense conflict, a claim may become fused with group identity. Criticism of policy, historical narrative, military strategy, or religious interpretation is then treated as criticism of the group’s right to exist. A bounded disagreement becomes betrayal, humiliation, desecration, or existential threat.

Research on sacred values suggests that some commitments are treated as categorically non-fungible and may resist ordinary material trade-offs (Tetlock et al. 2000; Ginges et al. 2007). Identity-fusion research describes conditions in which personal and group identity become highly aligned, potentially increasing willingness to undertake extreme pro-group action (Swann et al. 2009).

These findings do not imply that sacred values are irrational or uniquely violent. They may also support restraint, solidarity, protection, and non-negotiable human rights. The relevant question is which permission structure is produced.

20.3 Enemy construction and correction-path contraction

When threat, identity, and sacred value combine, the opponent may be represented as morally homogeneous, permanently hostile, incapable of correction, or outside the normal protection register. Its evidence is discounted because of identity. Defensive actions confirm hostility. Conciliatory actions may be interpreted as deception. The ontology becomes self-reinforcing.

Conflict also narrows time, information, and acceptable dissent. Emergency authority may centralise decisions, reduce evidence thresholds, classify information, and restrict public review. These changes may be necessary under genuine emergency. They become dangerous when temporary asymmetry loses expiry, independent review, restoration, or policy-level reopening.

20.4 Candidate escalation sequence

Threat and Uncertainty→Intersubjective Intensification→Identity–Claim Fusion→Sacralisation→Enemy Construction→Correction-Path Contraction→ADM Concentration→Escalation Permission.

(57)

Equation 55 is a candidate escalation sequence, not an estimated causal model. Its components may be absent, interrupted, reversed, or outweighed by security, economic, institutional, diplomatic, and leadership variables. The sequence identifies transitions for investigation; it does not predict war from identity, religion, sacred value, or intersubjective intensity alone.

Once escalation begins, each side may generate evidence confirming the other’s ontology:

Action→Counteraction→Threat Confirmation→Stronger Permission→Further Action.

(58)

This repeated-game structure becomes CEP-consistent only where a closure regime is established, actors possess continuation incentives, deviation is costly, expectations are sufficiently shared, and genuine de-escalation opportunities fail.

21. Technology as an Ontology Amplifier

Technology does not eliminate ontology. It materialises it. Every technical system must decide which objects are represented, which variables matter, which outcomes are optimised, which failures are detectable, and which actions are available. These decisions may remain implicit in data schemas, sensors, models, interfaces, thresholds, and automation rules.

Governing Ontology+Representation+Execution Capacity→Material Consequence.

(59)

21.1 Amplification without comprehensive understanding

A system need not possess a comprehensive understanding of the world it transforms. It may operate through reliable local prediction, constrained optimisation, pattern recognition, and procedural automation. The resulting action may still be powerful.

Danger increases where the model’s boundary disappears, the system controls consequential permission, and correction authority remains weak.

21.2 Local correctness and global incoherence

The Haas–Unger comparison supplies a useful structural insight:

Local Component Validity⇏Global Decision-Regime Coherence.

(60)

An AI workflow may contain an accurate model, valid policy, authorised operator, lawful database, and formal appeal channel while the integrated inference remains defective. Each local component passes its own test. No actor owns the combined ontology.

21.3 Medium and interface

Technology amplifies through representation as well as action. A system may express uncertainty internally while presenting operators with a binary label, ranked list, red warning, or default recommendation. The operator encounters the interface, not the complete model state.

Same Underlying Output+Different Interface→Different Permission Effect.

(61)

Interface design is therefore part of governance rather than a downstream usability concern.

21.4 Decision time and correction time

Let T_d denote decision or deployment time and T_c substantive correction time. The following is a conceptual timing-risk condition:

Td<Tc.

(62)

The system can generate, distribute, classify, restrict, or commit before the correction path can reconstruct the decision. The risk grows where action is scalable, automated, networked, irreversible, or difficult to attribute.

21.5 Execution–correction ratio

A candidate governance indicator is:

ECR=Execution CapacityOperative Correction Authority.

(63)

A rising ratio suggests that the system’s capacity to produce consequence is growing faster than the institution’s capacity to reopen and reverse that consequence. The ratio is conceptual rather than validated.

21.6 Automation bias and authority transfer

Research on automation bias shows that decision-makers may over-rely on automated recommendations and fail to detect errors, especially under workload or perceived system competence (Skitka, Mosier, and Burdick 1999; Parasuraman and Manzey 2010).

The problem is not simply excessive trust. Automation may transfer authority because the output appears precise, consistent, data-based, and institutionally certified. The operator may remain formally responsible while lacking practical capacity to reconstruct the model’s inference.

21.7 Technical lock-in

Once an ontology is encoded in data, APIs, procurement, training, and institutional reporting, revision becomes costly. A category may survive after its original justification weakens because every downstream system expects it, historical data depend on it, and institutional comparison would otherwise break.

This is technical path dependence. It must not be called CEP automatically. CEP requires additional evidence concerning strategic incentives, deviation costs, shared expectations, and failed correction opportunities.

Artificial intelligence intensifies every issue developed so far because it combines high formal performance, flexible representation, rapid execution, cross-domain integration, and increasingly agentic action. The central AI-governance question is not merely “Can the model reason?” It is:

Can justified evidence alter what the sociotechnical system is permitted to do?

Part VII — AI Governance

22. The Functional AI Analogue

The human–AI comparison developed in this article is functional rather than psychological. An artificial-intelligence system does not need human embodiment, autobiographical identity, social belonging, biological ontogenesis, or innate cognitive duality in order to reproduce a structural gap between formal capacity and governance reliability.

The relevant question is not whether an AI system possesses the same inner states as a human actor. It is whether the sociotechnical system displays comparable functional separations among capability, contextual realisation, directional use, evaluation, authority, correction, and permission.

22.1 The AI system is not the model alone

A deployed AI system should be represented as:

AIS={Mod,Prov,Dep,Wf,Eval,Op,Aff,Auth},

(64)

where Mod is the model; Prov the provider; Dep the deployer; Wf the workflow; Eval the evaluator; Op the human or automated operator; Aff the affected person or population; and Auth decision and correction authority.

The model is one component of the governed object. A harmful or defective outcome may originate in training data, system instructions, retrieval, tool permissions, workflow design, user interface, objective definition, evaluator policy, deployment incentives, or authority allocation. Attributing the complete outcome to “the model” may conceal the institutional decisions that made it consequential.

22.2 Capability realisation is contextual

A model does not express all available capabilities in every interaction. Realisation depends on prompt, system message, context window, retrieved material, tool access, sampling configuration, policy classifier, interaction history, and interface constraints.

Architecture and Representation→Training and Conditioning→Contextual Capability Realisation→Policy-Directed Utilisation→Embedded Task Ontology→Permission.

(65)

This sequence does not imply that model architecture determines every downstream state. Each transition contains organisational and technical choices.

22.3 AI formal capacity

In the AI context, formal capacity denotes demonstrated or reproducibly elicitable task competence under specified system conditions. It does not denote human-like comprehension, a unified latent faculty, moral agency, or context-independent general ability.

AI formal capacity may appear through mathematical reasoning, code generation, planning, argument reconstruction, classification, causal explanation, simulation, or tool-mediated execution. Benchmark performance may provide evidence concerning some of these capacities. It does not establish behaviour under conflicting objectives, ambiguous authority, incomplete evidence, user dependency, irreversible tools, or strong deployment incentives.

Benchmark Capability⇏Deployment Reliability.

(66)

22.4 Directional utilisation

AI directional utilisation is selected through training distribution, instruction hierarchy, objective, policy, evaluator, retrieval corpus, tool affordance, and interface. A system may be directed toward compliance, user satisfaction, refusal, prediction, persuasion, task completion, risk minimisation, or institutional throughput.

These directions may conflict. A system may generate sophisticated reasoning while using that reasoning primarily to preserve the instruction, requested objective, policy boundary, or deployer’s workflow. The existence of reasoning-like explanation does not establish that the reasoning can reopen those directing structures.

22.5 Corrective utilisation

An AI system displays a functional analogue of corrective utilisation only where its evaluation can reach the layer generating the defect. A model may state that the prompt is underspecified, objective is harmful, data are incomplete, policy conflicts with the user’s need, or task ontology excludes the affected party. This linguistic recognition is not yet correction.

Self-Critique≠Correction Sovereignty.

(67)

A configured system possesses no operative correction authority merely because it can describe why an action should not occur.

22.6 Agency is not governance authority

Agentic capability concerns the capacity to plan, call tools, maintain state, pursue multi-step objectives, and alter an external environment. Governance authority concerns legitimate control over objectives, permissions, scope, escalation, and reversal.

Agency⇏Legitimate Authority.

(68)

Increasing agentic capacity without corresponding correction authority raises the execution–correction imbalance identified in Section 21.

22.7 Human oversight is not one function

“Human oversight” may refer to observation, confirmation, exception handling, approval, appeal, intervention, policy design, or rollback authority. These functions should not be collapsed.

A human who may inspect an output but cannot delay, alter, restrict, or reverse the action does not possess operative oversight. A human who formally approves hundreds of automated decisions without time, evidence, or independent authority may serve as a procedural signature rather than a corrective agent. Research on government algorithms and public-sector human–AI interaction supports caution against treating nominal human involvement as a sufficient governance condition (Green 2022; Alon-Barkat and Busuioc 2023).

22.8 Functional mapping

Table 2. Functional Mapping between Human-Institutional and AI-System Layers

Human or institutional layer	AI-system analogue
Formal capacity	Demonstrated model and system competence under specified conditions
Partial realisation	Context-, prompt-, tool-, and workflow-sensitive performance
Directional utilisation	Objective, policy, evaluator, retrieval, interface, and proxy
Governing ontology	Task ontology and deployment problem model
Medium discipline	Interface, label, explanation, ranking, and action representation
Corrective utilisation	Evaluation capable of reopening the generating layer
Correction Sovereignty	Authority capable of changing permission
Governance reliability	Preserved correction under operational pressure

22.9 The affected party

The affected person is not merely an end user. A system may affect persons who never interact with it, cannot inspect its output, do not know it was used, or lack contractual standing with the provider. The governance unit must distinguish user, operator, decision subject, beneficiary, burden bearer, and correction claimant.

Failure to identify the affected party may produce need-profile substitution: the deployer’s efficiency objective is treated as though it were the decision subject’s need.

The article does not claim that an AI model understands authority as a human does, possesses moral agency, experiences socialisation, or consciously inhabits an ontology. It claims that a sociotechnical AI regime can display functional patterns in which capability is high, directional use is constrained, criticism is generated, authority is elsewhere, and correction fails to reach permission.

23. Embedded Task Ontologies and Evaluation Closure

Every consequential AI system operates inside a task ontology. The ontology may be explicit in a technical specification or distributed across data, labels, system instructions, workflow, interface, policy, and evaluator practice.

23.1 Task ontology

The article defines:

TO={Ent,Rel,Obj,Harm,Fail,Act,Med},

(69)

where Ent denotes recognised entities; Rel relations; Obj the operative objective; Harm the harm ontology; Fail recognised failure classes; Act permitted actions; and Med medium and representation.

The evaluator can detect only failures for which it possesses categories. An omitted entity cannot enter the decision except indirectly. A system may represent individuals and outcomes while omitting institutional causation or burden transfer. An objective may be explicit in a loss function or implicit in workflow incentives. The harm ontology may include physical injury, legal violation, financial loss, discrimination, privacy invasion, misinformation, or security compromise while omitting dependency, humiliation, long-term institutional erosion, correction cost, or burden transfer.

23.2 Permitted actions

The same model can occupy radically different governance states depending on whether it may answer, recommend, rank, write a record, send a message, approve, deny, call a tool, transfer money, or modify infrastructure. Model capability does not determine action permission.

23.3 Medium and representation

The medium determines how the result enters human and institutional cognition. The same underlying judgment may appear as a confidence interval, binary classification, recommendation, default, risk score, ranked list, alert, or automatic action.

Same Inference+Different Representation→Different Permission Effect.

(70)

Medium therefore belongs inside the task ontology, not outside it.

23.4 Inherited task ontology

AI systems frequently enter decision regimes that existed before the model. They inherit institutional categories, historical labels, eligibility rules, documentation practices, proxy variables, and prior burden distributions. The model may amplify a pre-existing problem rather than create it.

The correct governance question is not only “Did the AI introduce an error?” It is also “Which unresolved decision structure did the AI operationalise?”

23.5 Benchmark validity and deployment legitimacy

A benchmark may be valid for a bounded measurement purpose. This does not establish that the measured capability should govern a consequential deployment.

Benchmark Validity≠Deployment Legitimacy.

(71)

Deployment legitimacy requires additional judgments concerning purpose, affected population, authority, error distribution, reversibility, alternatives, and remedy.

23.6 Evaluation closure

Evaluation closure occurs where evaluation remains confined to the ontology and objective of the system being evaluated. The evaluator may ask whether the model performs the assigned task, complies with policy, improves the target metric, and keeps recognised failure classes below threshold. It may not ask whether the task is properly constituted, the objective legitimate, the metric representative, the affected party correctly identified, or the action permitted at all.

A closed evaluation can be technically sophisticated. Its closure concerns jurisdiction.

23.7 Shared-ontology closure

Provider, deployer, and evaluator may share the same problem model. Agreement among them may then reflect common prior structure rather than independent validation.

Evaluator Independence≠Organisational Separation Alone.

(72)

Meaningful independence may require authority to select tests, access adverse evidence, redefine failure, challenge the task ontology, publish or escalate negative findings, and affect permission.

23.8 Proxy capture

A proxy becomes dangerous where it moves from measurement instrument to governing object:

Proxy→Target→Permission.

(73)

Examples include treating engagement as value, complaint reduction as satisfaction, throughput as care quality, benchmark score as intelligence, or policy compliance as safety. Campbell’s law, audit-effects literature, and Goodhart-type analyses provide established antecedents for the corruption and displacement risks created when metrics become consequential targets (Campbell 1979; Strathern 1997; Manheim and Garrabrant 2018).

The article’s stronger concern is ontological: the proxy may replace the object before metric degradation becomes visible.

23.9 Core and Shell

An AI system may exhibit a stable Shell: compliant formatting, disclaimers, safe language, procedural explanation, and policy-consistent presentation. Its Core reasoning may remain unstable.

Core/Shell instability occurs where surface compliance conceals contradictory causal reasoning, authority confusion, object substitution, or inconsistent treatment of the same principle. A Shell evaluator may reward the appearance of governance while failing to inspect the reasoning structure generating action.

23.10 Versioned task ontology

A task ontology should be versioned. Material changes in model, dataset, objective, population, workflow, authority, tool access, or time-sensitive environment may create a new governance object. Approval of one configuration should not transfer automatically to another.

Task-ontology review should be triggered where recurring errors cannot be resolved at output level, affected-party evidence reveals an omitted harm, model performance is strong but real-world outcome poor, evaluator disagreement concerns the object rather than the score, proxy improvement accompanies need deterioration, or the action class becomes more consequential.

Stable evaluator agreement does not prove evaluation closure. Agreement may reflect strong evidence, mature measurement, or robust performance. Closure requires evidence that relevant ontology-level questions lack an authorised route into evaluation or permission.

24. From Output Correction to Governance Correction

AI systems often generate a dense layer of evaluative activity: benchmark results, safety scores, audits, red-team findings, incident reports, human reviews, model cards, monitoring dashboards, and self-critical outputs. These artefacts may be useful. They do not automatically constitute correction.

24.1 Complete correction chain

Criticism→Feedback→Review→Judgment→Gate Change→Implemented State Change→Verification and Restoration.

(74)

Criticism is an objection, anomaly, dissenting interpretation, failure signal, or claim that the current state is inadequate. Feedback transmits information into a receiving process. Review is substantive examination by a competent forum. Judgment is a reasoned conclusion concerning evidence, object, failure, and appropriate response.

A gate or authorised decision change constitutes decision correction. A verifiable change in system state, workflow, access, model version, authority, or deployment constitutes operative correction. Correction of the future state does not erase prior consequence; where appropriate, correction may also require record amendment, reinstatement, notification, compensation, re-evaluation, or recurrence prevention.

24.2 AI correction hierarchy

The seven-level general hierarchy introduced in Section 11 distinguishes the objects of correction. For operational AI analysis, closely related levels are consolidated into six implementation classes:

output correction;
model correction;
evaluation correction;
workflow and representation correction;
task-ontology correction;
governance and authority correction.

The consolidation is practical rather than theoretical. Representation is grouped with workflow, and authority correction with governance correction. This does not erase the distinctions required for causal diagnosis.

The governing rule is:

Correction Depth≥Failure-Generation Depth.

(75)

A local output error should not trigger unnecessary architectural redesign. A structural ontology defect should not be treated as an isolated output error.

24.3 Self-revision and governance correction

Model-level self-revision methods may improve outputs under some conditions. They do not constitute governance correction unless the result can alter workflow, ontology, authority, or permission.

A model criticising its own answer is not equivalent to an institution capable of restricting or rolling back deployment.

24.4 Symbolic governance and soft closure

Symbolic governance consists of artefacts that express responsibility or control without a reliable connection to permission: an alert that cannot trigger a gate, an audit that cannot affect deployment, a safety board whose findings are advisory only, a HOLD status without technical enforcement, a reviewer without reversal authority, or a rollback policy unsupported by versioned infrastructure.

Such artefacts may preserve evidence and improve communication. They become misleading where their existence is treated as proof that the system can correct itself.

Soft closure occurs where criticism, feedback, transparency, and review remain available while the operative state is protected:

Criticism→Feedback→Review↛Operative Correction.

(76)

The institution may appear highly open. Its closure lies at the point of consequence.

24.5 Human oversight conditions

Human oversight is corrective only where the human possesses relevant evidence, adequate time, competence, independence, action authority, technical means of intervention, protection from retaliation, and a traceable responsibility path (Green 2022; Alon-Barkat and Busuioc 2023; European Parliament and Council of the European Union 2024).

The presence of a human in the workflow does not satisfy these conditions automatically.

24.6 Evaluator disagreement

Evaluator disagreement is not necessarily noise. It may indicate ambiguity, ontology conflict, threshold instability, or genuine value pluralism. A governance architecture should distinguish disagreement within one metric, disagreement between metrics, disagreement concerning relevant harm, and disagreement concerning the legitimate objective. High-level disagreement may justify HOLD rather than forced averaging.

24.7 Correction of correction

A governance system must evaluate its own correction mechanism. Are appeals accessible? Are reversals implemented? Are false HOLDs monitored? Does ROLLBACK occur too late? Are evaluators overfitted? Does the correction process create new burden? Can affected parties challenge the ontology used by the review body?

A correction mechanism that cannot itself be corrected becomes a new source of closure.

Candidate measures include evaluation-to-gate conversion rate, high-risk signal-to-HOLD rate, serious violation-to-ROLLBACK rate, evaluator disagreement-to-escalation rate, formal reversal-to-implementation rate, implementation-to-restoration rate, unresolved authority-ambiguity rate, false-HOLD rate, missed-critical-failure rate, and time from signal to operative correction. These are candidate metrics, not validated measures of governance reliability.

25. LoopGuard-AI and Permission-State Governance

LoopGuard-AI is an advanced candidate permission-state architecture. The present article specifies its conceptual objects, decision sequence, gate grammar, audit requirements, and defeat conditions. It does not establish calibrated thresholds, comparative superiority, production reliability, feasible rollback across deployment classes, or demonstrated improvement over existing governance mechanisms.

25.1 Governing proposition

Evaluation becomes governance only when admissible evidence crosses a justified threshold, reaches a competent authority, produces a proportionate gate, and changes the permission state of the governed object.

The two-layer corrective principle operates here as a permission invariant. Objective-critical reason is not converted into a model-generated moral score, and epistemological priority is not reduced to generic uncertainty disclosure. The invariant requires that no task ontology, objective, benchmark, policy, evaluator, or deployment assumption become immune from the same evidence-to-authority path used to correct outputs. A gate is therefore governance-relevant only if it can be reopened at the highest level generating the defect—including ontology and purpose—and only if the authorised change can be implemented, reversed, and reviewed.

Signal→Metric→Threshold→Authority→Gate→Consequence

(77)

A signal may arise from model output, red-team test, evaluator, user report, affected-party evidence, drift monitor, policy conflict, tool trace, security incident, or performance anomaly. The metric specifies how the signal is represented and compared. The threshold determines when evidence becomes decision-bearing. A competent authority must possess jurisdiction, evidence access, mandate, technical ability to alter the state, and responsibility for implementation. The gate translates judgment into permission. The consequence is the verifiable state change.

25.2 Governed permission object

Permission should attach to a versioned configuration rather than to “the model” in the abstract:

Π=(Mod,Data,Use,Pop,Wf,Auth,Time).

(78)

The tuple identifies model and version, data and retrieval configuration, intended use, affected population, workflow and tools, authority level, and time period or review horizon. Material change in one component may create a new governance object.

25.3 Gate grammar

LoopGuard-AI proposes four principal states:

G={SHIP,RESTRICT,HOLD,ROLLBACK}.

(79)

SHIP permits deployment, continuation, or execution under defined conditions. It does not mean zero risk, universal validity, or permanent approval. A SHIP state should specify scope, population, monitoring, assumptions, expiry, and rollback conditions.

RESTRICT permits continuation only under limitation. Restrictions may concern scope, autonomy, tool access, rate, population, human confirmation, logging, or permission period.

HOLD pauses permission pending additional evidence, review, authority clarification, testing, or escalation. HOLD is not rejection. Candidate triggers include high risk with incomplete evidence, evaluator disagreement, unclear authority, limited reversibility, suspected drift, policy conflict, incomplete audit record, or possible Core/Shell instability.

ROLLBACK reverses, withdraws, disables, downgrades, or returns the system to a safer prior state. Candidate triggers include severe policy violation, high-impact unsafe action, persistent drift, failed release, unacceptable irreversibility, unstable governance behaviour, or inability to reconstruct and audit the decision.

25.4 Gate derivation and representation condition

The gate should be derived from a structured evidence profile rather than one unqualified aggregate score. Relevant fields include failure class, severity, confidence, affected party, reversibility, authority, recurrence, evaluator agreement, drift, policy conflict, and correction level.

Each gate package should also identify how the evidence and decision are represented to operators, authorities, affected parties, and reviewers. The record should state which uncertainty is visible, which uncertainty is suppressed, whether the interface encourages over-reliance, and whether a change in representation could change the operative decision.

25.5 Audit-ready decision record

A governance record should include, where applicable, event identifier, timestamp, system and version, triggering condition, output and action trace, tool-call trace, evaluator versions, raw and normalised evaluator results, uncertainty, policy version, triggered thresholds, authority assessment, reversibility assessment, drift assessment, gate decision, rationale, escalation, override, implementation status, and restoration status.

A decision that cannot be reconstructed remains weak as governance evidence.

25.6 Replayability and rollback readiness

Where technically feasible, the architecture should support replay. Replay does not guarantee identical model output where stochastic generation is present. It should preserve enough information to determine why the decision differed, which component changed, and whether gate logic operated consistently.

Formal rollback authority is insufficient where reversal is technically or organisationally impossible. Rollback readiness may require version control, retained prior models, alternative workflows, data and configuration snapshots, stop authority, downstream containment, and communication procedures.

25.7 Correction-owner assignment

Every gate must have an identifiable owner. The owner may differ across model, workflow, policy, ontology, and deployment. Distributed responsibility is not itself a failure. Failure occurs where no actor owns the complete correction required.

25.8 Independent evaluation and confidentiality

Evaluation is independent only where the evaluator can do more than issue an opinion. Relevant powers may include test selection, adverse-evidence access, escalation, protected reporting, and permission effect. Complete independence may not always be feasible; the required level should be proportionate to consequence and conflict of interest.

Confidentiality does not make governance impossible. Cleared review, compartmented evidence access, confidential appeal, protected audit, and aggregate reporting can preserve reviewability. Universal public disclosure is not required. Unreviewable secrecy is insufficient.

25.9 Relation to existing frameworks

Existing frameworks recognise multiple functions relevant to risk management, monitoring, documentation, corrective action, human oversight, contestation, restriction, and withdrawal (National Institute of Standards and Technology 2023; Autio et al. 2024; Organisation for Economic Co-operation and Development 2024; European Parliament and Council of the European Union 2024).

LoopGuard-AI proposes one way of integrating these functions around explicit versioned permission states. The present article does not establish that this integration is necessary or superior.

The architecture should therefore be evaluated as a candidate integration layer rather than as a replacement for law, standards, technical safety research, organisational governance, or sector-specific expertise.

25.10 Candidate metrics and CEP sensitivity

Possible metrics include evidence completeness, explainable-gate rate, audit reconstruction rate, gate reproducibility, override-trace completeness, false-HOLD rate, missed-critical-failure rate, time to correction, reversal implementation rate, restoration completion rate, and unresolved policy-conflict rate.

LoopGuard-AI may flag CEP-sensitive instability where the same weak decision is repeatedly justified, policy compliance substitutes for conceptual discrimination, evaluators stabilise one answer despite unresolved ontology conflict, or the regime rewards closure more reliably than correction.

A CEP flag is not a gate by itself. It is a higher-order signal requiring separate evidence concerning closure, incentives, deviation cost, expectations, and missed correction opportunities.

The minimum constructive claim is:

A consequential AI system is not governed merely because it is evaluated. It is governed only where justified evaluation can change an explicit, enforceable, reviewable, and reversible permission state.

Part VIII — Research Discipline

26. Falsifiability, Measurement, and Rival Explanations

A theory of governance failure is useful only if it can fail. This article treats falsifiability as a methodological discipline rather than a single binary test for every social-scientific construct (Popper 1959). Conceptual richness creates a particular danger: every anomaly can be redescribed as another manifestation of the framework. The theory then appears comprehensive because no evidence can count against it.

Conceptual Coherence≠Measurability≠Explanatory Superiority.

(80)

26.1 Modular theory structure

The integrated theory may be represented as:

T={TFC,TDU,TDIC,TPIC,TOPI,TOA,TEP,TSLI,TDCA,TCVA,TCEP,TAI,TLG}.

(81)

Here T_EP denotes the two-layer epistemological-priority principle and T_SLI the scientific-legitimation-interface hypothesis. Failure of one module does not automatically destroy every other module. DIC may fail while the capacity–correction distinction survives. The scientific-legitimation-interface hypothesis may fail in a specified domain while epistemological priority remains a normative governance requirement. CEP may fail in a specified case while DCA remains supported. LoopGuard-AI may prove operationally inferior while the need for evaluation-to-permission conversion remains valid.

26.2 Permitted research outcomes

A serious test may produce:

CONFIRM — the construct performs as predicted across independent evidence;
NARROW — it applies only under a smaller set of conditions;
SIMPLIFY — a simpler established concept explains the evidence equally well;
RECLASSIFY — the phenomenon belongs at a different level or under another construct;
SPLIT — one construct contains several empirically distinct mechanisms;
REJECT — the proposed relation is unsupported or contradicted.

The ten-case comparative test already produced a legitimate negative result: the binary profile-cluster model was not supported as the final classification of the selected corpus.

26.3 Measuring formal capacity and utilisation

Formal capacity should be measured using domain-appropriate performance rather than reputation alone. Possible indicators include abstract reasoning, counterfactual analysis, formal proof, complex modelling, source criticism, mechanism reconstruction, and structured comparison. Mathematical form should not be treated as the only formal capacity.

The multidimensional utilisation model requires an explicit codebook, bounded corpus, domain and genre controls, counterevidence coding, multiple coders, and reliability testing. The dimensions must show discriminant validity. If boundary discipline and mechanism specification cannot be distinguished reliably, they should be merged or redefined.

26.4 Measuring corrective symmetry and institutional conversion

Corrective symmetry should be evaluated at a specified question level. Candidate indicators include explicit rival hypotheses, treatment of negative evidence, revision of prior claims, application of equal evidence thresholds, and recognition of failure conditions for the governing framework.

Institutional conversion requires comparison between diagnosis and operative consequence. Was a responsible authority identified? Did the recommendation alter a decision? Was the alteration implemented? Was the generating rule changed? Was restoration completed? Did similar failures decline?

Conversion must be assessed relative to role. A scholar without administrative authority should not be judged by the same implementation criterion as a regulator or deployer.

26.5 Measuring DCA, CVA, and CEP

DCA requires evidence across standing, evidence, threshold, authority, implementation, time, policy, and representation. One failed appeal is insufficient.

CVA requires longitudinal and denominator-sensitive evidence concerning eligible corrective participants, attempted voice, substantive review, correction outcome, later voice behaviour, reassignment, silence, and exit. It is defeated where turnover is explained better by pay, workload, ordinary mobility, or unrelated organisational change.

CEP requires evidence beyond persistence: closure regime, actor-specific payoffs, deviation costs, expectations concerning others, and genuine correction opportunities. If these are absent, the analysis should remain at pattern, closure, or path dependence.

26.6 Measuring AI correction

Candidate AI-governance outcomes include evaluation-to-gate conversion, gate-to-state implementation, rollback feasibility, restoration, recurrence, evaluator independence, and affected-party access. Model performance alone is not the dependent variable.

26.7 Research designs

Relevant designs include:

comparative textual designs using bounded public corpora and explicit counterevidence;
within-profile designs comparing the same actor across domain, genre, role, and period;
institutional episode designs tracing one permission object from evidence through authority to consequence and correction;
longitudinal voice designs measuring whether failed correction attempts predict later silence or exit;
AI intervention designs varying evaluator access, authority, interface, threshold, or gate enforcement;
simulation designs testing classification consistency before prevalence claims.

Confirmatory studies should preregister sample frame, case-selection rule, unit boundaries, codebook, primary variables, outcome definitions, exclusion rules, missing-data rules, rival explanations, analysis method, and falsification conditions (Shadish, Cook, and Campbell 2002).

26.8 Construct validity and reliability

Validity belongs to a specified inference and use, not to a label considered in the abstract. The empirical programme should distinguish concept formation, specification, content representation, unitisation reliability, component reliability, convergent validity, discriminant validity, incremental value, intervention evidence, and transfer (Adcock and Collier 2001; Campbell and Fiske 1959; Messick 1995; Krippendorff 2004; American Educational Research Association, American Psychological Association, and National Council on Measurement in Education 2014).

Researchers should report agreement, disagreement, adjudication, and codebook revision. High disagreement may indicate poor training, ambiguous evidence, or a defective construct. It should not be concealed through forced consensus.

26.9 Rival explanations

Principal rivals include discipline, genre, role, ideology, identity, professional incentive, organisational silence, principal–agent failure, path dependence, hierarchy, resource scarcity, legal constraint, automation bias, and ordinary measurement error.

The integrated theory earns value only where it improves classification, prediction, explanation, or intervention after these rivals are included.

26.10 False precision

Formal notation may preserve distinctions while creating an appearance of mathematical validation. The equations in this article are schematic. They do not establish effect size, probability, equilibrium existence, or causal identification. A construct should not receive a scalar score merely because one can be written.

26.11 Module-specific defeat conditions

DIC should be removed if it fails to explain variance beyond discipline, genre, role, socialisation, medium, institution, ideology, and incentive, or if proposed tendencies lack developmental stability, within-person consistency, or predictive value.

OPI should be narrowed or rejected where coders cannot distinguish projection from legitimate developmental explanation, developmental vocabulary produces excessive false positives, or mechanism and boundary criteria explain the same material more simply. OPI must never classify a claim solely because it uses words such as growth, development, evolution, stage, or maturation.

The non-substitutable claim for epistemological priority should be narrowed or rejected if decision regimes can be shown to preserve stable, consequential, pressure-resistant correction while their operative ontologies remain categorically immune from epistemological reopening. Its sufficiency is already denied: evidence that epistemological openness fails without authority, implementation, reversibility, or restoration supports rather than refutes the article’s bounded formulation.

The scientific-legitimation-interface hypothesis should be rejected in a case where no authority transfer occurs, where each transition from bounded claim to ontology and permission is independently justified, or where the governing permission does not depend materially on scientific prestige. Collective ontogenesis projection should remain outside the load-bearing theory unless independent coding establishes cross-scale transfer beyond ordinary developmental language, disciplinary convention, and narrative convenience.

LoopGuard-AI should be narrowed, redesigned, or rejected if its gates do not improve decisions over existing controls, false HOLD or ROLLBACK costs are excessive, thresholds cannot be calibrated, audit records create prohibitive burden, authority mapping fails in real organisations, rollback is usually infeasible, or the architecture becomes symbolic governance.

27. Limitations and Claim-Maturity Map

The manuscript’s theoretical breadth creates both its contribution and its principal limitation. It connects cognition, epistemology, ontology, institutional authority, conflict, and AI governance. No single empirical study could validate the full architecture.

The ten-profile corpus is purposive, non-random, culturally concentrated, and unevenly accessible. It can reject the initial binary classification within the studied design. It cannot establish the distribution of utilisation dimensions in a population.

Public work is not private cognition. Published text may reflect editorial constraint, strategic presentation, audience, role, legal risk, and translation. No psychological diagnosis follows.

The manuscript identifies candidate mechanisms but does not establish their complete causal sequence. PIC dependence may correlate with reduced constitutive criticism without causing it. DCA may coexist with organisational silence while another mechanism generates both.

Structural resemblance across person, institution, civilisation, and AI system does not establish shared origin or ontology. The human–AI analogue is functional.

The article also contains normative commitments. It assumes that consequential decision regimes should preserve contestability, proportionality, reversibility, correction, and restoration. These commitments are argued but not derived from descriptive evidence alone.

The two-layer corrective principle is a normative-operational synthesis. The article does not establish that Horkheimer and Adorno formulated epistemological priority, that the historical West uniquely embodied it, or that every scientific theory becomes an institutional legitimation interface. “Exclusive” means non-substitutable within correction architecture; it does not mean sufficient without institutional embodiment. Collective ontogenesis projection remains an open stress-test hypothesis, not a restored foundation for DIC.

The notation is conceptual. No equation should be interpreted as an estimated causal model, validated index, or formal equilibrium proof.

The ontological attractors are ideal types. DIC remains speculative and non-load-bearing. OPI may overclassify legitimate historical or developmental explanation. Correction failure is often difficult to observe because evidence may exclude abandoned complaints, private dissent, classified review, informal correction, and successful cases that generate no controversy. AI outcomes are distributed across many layers, making both model-centred and excessively diffuse attribution risky.

LoopGuard-AI has not been validated in production. The architecture has not demonstrated threshold calibration, comparative superiority, operational cost, real rollback effectiveness, or governance reliability under deployment pressure.

Table 3. Final Claim-Maturity Map

Claim or construct	Final maturity status
Formal capacity differs from realised use	Strong conceptual proposition
Realisation is domain- and context-sensitive	Externally supported background
Five-stage ladder	Advanced conceptual architecture
Directional utilisation is multidimensional	Supported by purposive comparison; not population-validated
Two stable profile clusters	Not supported in the ten-case corpus
Boundary orientation predicts non-conversion	Not supported
Social-symbolic orientation predicts OPI	Not supported
Prior explanatory orientations	Possible non-exclusive tendencies
DIC as innate duality	Speculative and non-load-bearing
PIC	Developed construct; causal role unvalidated
OPI	Diagnostic candidate requiring reliability testing
Two ontological attractors	Developed regime-level ideal types
Objective-critical reason	External normative antecedent; not identical to epistemological priority
Epistemological priority	Non-substitutable governance condition; normative-operational proposition
Two-layer corrective principle	Developed synthesis requiring institutional validation
Scientific-legitimation interface	Open institutional mechanism hypothesis
Collective ontogenesis projection	Open stress-test hypothesis; OPI diagnostic only
ADM/CIV	Advanced functional architecture
Three Sovereignties	Advanced institutional distinction
Protected Contestability	Normative-operational proposal
Affective-Jurisdiction Failure	Operationalisable candidate construct
Sentimental Veto classifier	Developed local classifier; external validation required
DCA	Developed and operationalisable
CVA	Open longitudinal mechanism hypothesis
CEP-consistent persistence	Downstream equilibrium-like hypothesis
Task ontology	Strong sociotechnical construct
AI correction hierarchy	Developed architecture
Evaluation-to-gate conversion	Operationalisable governance proposition
SHIP / RESTRICT / HOLD / ROLLBACK	Proposed permission grammar
LoopGuard-AI	Advanced candidate architecture; empirically unvalidated

The article should therefore be classified as:

an advanced candidate analytical theory and research programme with an unvalidated governance architecture.

Its conceptual distinctions are mature enough to be criticised, operationalised, compared, narrowed, and rejected. Its empirical maturity remains uneven.

28. Conclusion — Intelligence after the Loss of Corrective Authority

Advanced intelligence is commonly recognised through what it can produce. It can abstract, calculate, model, predict, narrate, classify, optimise, and explain. These capacities matter. They do not establish that intelligence governs.

The central distinction of this article is:

Formal Capacity≠Governance Reliability.

Between them lie several independent transitions:

Formal Capacity→Partial Realisation→Directional Utilisation→Corrective Utilisation→Governance Reliability.

Each transition can fail while the preceding capacity remains visible.

28.1 The failed binary and the stronger result

The article began with a candidate binary distinction between boundary-oriented reasoning and developmental or social-symbolic reasoning. The ten-profile comparison did not confirm that classification as the final result of this corpus.

Boundary discipline, mechanism, development, representation, and social-symbolic intelligibility appeared in different combinations across mathematics, physics, biology, philosophy, aesthetics, history, political sociology, religion critique, and critical theory. The predicted relation between boundary discipline and institutional non-conversion was contradicted by major cases. The predicted relation between social-symbolic interpretation and ontogenesis projection was not established.

This negative result prevents the article from explaining governance failure by assigning persons to two intellectual types.

28.2 Multidimensional formal utilisation

Formal utilisation is directional but multidimensional. A reasoning regime may preserve strong evidence boundaries yet fail to map authority. It may reconstruct public meaning while preserving rigorous mechanism. It may explain historical development without projecting ontogenesis. It may criticise external institutions while protecting its own governing framework. It may produce an accurate diagnosis without possessing authority to change permission.

No one intellectual strength guarantees the complete correction chain.

28.3 Corrective intelligence

The article proposes a stronger conception of intelligence:

CI=FC+CA+OA+Rev+Learn,

(82)

where FC is formal capacity; CA corrective access; OA operative authority; Rev reversibility and restoration; and Learn institutional learning.

The expression is schematic. It identifies missing components in ordinary capacity-centred accounts.

28.4 Governance reliability

Governance reliability is:

the stable capacity of a decision regime to preserve consequential correction under the conditions most likely to suppress it.

Those conditions include uncertainty, time pressure, incentive conflict, authority ambiguity, institutional self-protection, identity threat, commercial pressure, and irreversible action. A system that corrects only where correction is inexpensive is not reliably governed.

28.5 Ontology and fragmented intelligence

Every consequential decision regime rests on an ontology. The ontology determines the entities, relations, harms, purposes, and correction possibilities the system can recognise. The central governance transition is:

Ontology→Permission.

Where that transition is hidden, technical capability may operationalise a defective world-picture with extraordinary efficiency.

Modern institutions distribute intelligence across specialists, administrators, affected persons, critics, evaluators, and technical systems. This distribution can preserve pluralism and prevent one institution from controlling the complete chain from knowledge to coercion. It can also leave the chain ownerless.

Scientific evidence may be strong. Historical interpretation may be sophisticated. Affected-party testimony may be credible. Administrative execution may be efficient. Yet no actor may possess authority over the complete inference from evidence to ontology, ontology to purpose, purpose to permission, permission to consequence, and consequence back to correction.

28.6 Criticism without authority

A system may appear open because it permits criticism, feedback, review, audit, transparency, and appeal. The decisive question is whether justified criticism can change the operative state. Where it cannot, openness is symbolic.

Correction failure also changes the population of participants. Actors who repeatedly encounter non-operative voice may move toward silence, accommodation, reduced participation, or exit. The institution then loses evidence, expertise, memory, and adversarial capacity. The resulting quiet may be misread as consensus.

Under stricter conditions, closure may become CEP-consistent. Relevant actors may continue because unilateral deviation is costly, others are expected to persist, correction attempts are anticipated to fail, and the arrangement remains locally rational. No conspiracy is required. A system can be strategically stable and collectively defective.

28.7 AI intensification

Artificial intelligence intensifies the problem because it integrates representation, evaluation, recommendation, and execution across domains previously separated by institutional boundaries. A model can combine scientific, legal, political, ethical, and administrative language while correction authority remains distributed among providers, deployers, evaluators, regulators, operators, and affected persons.

The system may act as one before it can correct itself as one.

AI systems may generate impressive self-criticism. They may identify uncertainty, bias, policy conflict, ontology failure, or harmful consequence. That capacity is valuable. It does not govern the system unless the criticism can reach an authority, change a gate, alter implementation, and support reversal or restoration.

Evaluation≠Governance

until:

Evaluation→Permission Change.

28.8 The two-layer corrective principle

The article’s positive solution is therefore two-layered.

At the level of essence, epistemological priority requires every consequence-producing ontology to remain provisional, justified, contestable, and reopenable. This is the non-substitutable condition: no amount of expertise, efficiency, consensus, transparency, or evaluation can produce governance reliability where the accepted ontology defines the limits of admissible criticism against itself.

At the level of intellectual genealogy, objective-critical reason supplies the authority to judge ends. Horkheimer and Adorno diagnose the failure that occurs when reason survives mainly as calculation, adaptation, administration, and control. The present article converts that diagnosis into a corrective constitution: ends remain answerable to reason because the ontologies through which they become permissions remain answerable to epistemology.

The complete relation is not:

objective reason equals epistemological priority.

It is:

objective reason authorises critique of ends; epistemological priority preserves the institutional conditions under which that critique can change ontology and permission.

In ADM–CIV relations, this allows administrative closure without ontological finality. ADM can coordinate action under uncertainty while CIV evidence retains standing, Correction Sovereignty remains operative, and permissions remain reversible. In AI governance, the same principle requires model, evaluator, task ontology, representation, objective, and deployment permission to remain connected to one auditable correction path.

The principle is necessary but not sufficient. Without authority, implementation, restoration, and learning, epistemological priority remains symbolic. Without epistemological priority, those mechanisms may become highly efficient instruments for preserving an ontology they are not authorised to question.

28.9 Final proposition

The article does not establish biological cognitive duality, a universal profile taxonomy, a complete causal theory of war, prevalence of institutional closure, equilibrium in every persistent institution, or validated superiority of LoopGuard-AI.

It establishes a structured research object:

the non-conversion of advanced formal capacity into stable corrective authority.

Formal reason does not govern merely because it can abstract, calculate, model, predict, explain, narrate, or optimise. It becomes governance-relevant only where it can expose the categories, purposes, representations, ontologies, authority relations, and permission states directing its own use to consequential correction.

Where formal capacity is socially canalised, institutionally filtered, and repeatedly stabilised without operative reversal, intelligence may become extraordinarily powerful while remaining structurally unable to govern the world it is increasingly able to transform.

The defining problem of advanced intelligence is therefore no longer capacity alone. It is whether intelligence capable of transforming the world remains institutionally authorised to correct the world-picture, medium, objective, authority structure, and permission regime directing that transformation.

References

Cognition, Development, and Formal Reason

Horkheimer, Max. 1947. Eclipse of Reason. New York: Oxford University Press.

Horkheimer, Max, and Theodor W. Adorno. 2002. Dialectic of Enlightenment: Philosophical Fragments. Edited by Gunzelin Schmid Noerr. Translated by Edmund Jephcott. Stanford, CA: Stanford University Press.

Kuhn, Deanna. 2008. Education for Thinking. Cambridge, MA: Harvard University Press.

Moore, John C., and Louis J. Rubbo. 2012. “Scientific Reasoning Abilities of Nonscience Majors in Physics-Based Courses.” Physical Review Special Topics—Physics Education Research 8 (1): 010106. https://doi.org/10.1103/PhysRevSTPER.8.010106.

Piaget, Jean. 1972. “Intellectual Evolution from Adolescence to Adulthood.” Human Development 15 (1): 1–12. https://doi.org/10.1159/000271225.

Epistemology, Ontology, and Sociology of Knowledge

Ben-Menahem, Yemima. 1997. “Historical Contingency.” Ratio 10 (2): 99–107.

Ben-Menahem, Yemima. 2018. Causation in Science. Princeton, NJ: Princeton University Press.

Berger, Peter L., and Thomas Luckmann. 1966. The Social Construction of Reality: A Treatise in the Sociology of Knowledge. Garden City, NY: Anchor Books.

Snow, C. P. 1959. The Two Cultures and the Scientific Revolution. Cambridge: Cambridge University Press.

Abbott, Andrew. 1988. The System of Professions: An Essay on the Division of Expert Labor. Chicago: University of Chicago Press.

Institutions, Authority, Voice, and Path Dependence

Abbott, Kenneth W., and Duncan Snidal. 2009. “The Governance Triangle: Regulatory Standards Institutions and the Shadow of the State.” In The Politics of Global Regulation, edited by Walter Mattli and Ngaire Woods, 44–88. Princeton, NJ: Princeton University Press.

Bovens, Mark. 2007. “Analysing and Assessing Accountability: A Conceptual Framework.” European Law Journal 13 (4): 447–68. https://doi.org/10.1111/j.1468-0386.2007.00378.x.

Detert, James R., and Ethan R. Burris. 2007. “Leadership Behavior and Employee Voice: Is the Door Really Open?” Academy of Management Journal 50 (4): 869–84. https://doi.org/10.5465/amj.2007.26279183.

Detert, James R., and Amy C. Edmondson. 2011. “Implicit Voice Theories: Taken-for-Granted Rules of Self-Censorship at Work.” Academy of Management Journal 54 (3): 461–88. https://doi.org/10.5465/amj.2011.61967925.

Edmondson, Amy C. 1999. “Psychological Safety and Learning Behavior in Work Teams.” Administrative Science Quarterly 44 (2): 350–83. https://doi.org/10.2307/2666999.

Hirschman, Albert O. 1970. Exit, Voice, and Loyalty: Responses to Decline in Firms, Organizations, and States. Cambridge, MA: Harvard University Press.

Kaminski, Margot E., and Jennifer M. Urban. 2021. “The Right to Contest AI.” Columbia Law Review 121 (7): 1957–2048.

Koppell, Jonathan G. S. 2005. “Pathologies of Accountability: ICANN and the Challenge of ‘Multiple Accountabilities Disorder.’” Public Administration Review 65 (1): 94–108. https://doi.org/10.1111/j.1540-6210.2005.00434.x.

Morrison, Elizabeth W. 2014. “Employee Voice and Silence.” Annual Review of Organizational Psychology and Organizational Behavior 1: 173–97. https://doi.org/10.1146/annurev-orgpsych-031413-091328.

Morrison, Elizabeth W., and Frances J. Milliken. 2000. “Organizational Silence: A Barrier to Change and Development in a Pluralistic World.” Academy of Management Review 25 (4): 706–25. https://doi.org/10.5465/amr.2000.3707697.

Pierson, Paul. 2000. “Increasing Returns, Path Dependence, and the Study of Politics.” American Political Science Review 94 (2): 251–67. https://doi.org/10.2307/2586011.

Sydow, Jörg, Georg Schreyögg, and Jochen Koch. 2009. “Organizational Path Dependence: Opening the Black Box.” Academy of Management Review 34 (4): 689–709. https://doi.org/10.5465/amr.34.4.zok689.

Uncertainty, Identity, Sacred Values, and Conflict

Gelfand, Michele J., Jana L. Raver, Lisa Nishii, Lisa M. Leslie, Janetta Lun, Beng Chong Lim, Lili Duan, et al. 2011. “Differences between Tight and Loose Cultures: A 33-Nation Study.” Science 332 (6033): 1100–1104. https://doi.org/10.1126/science.1197754.

Ginges, Jeremy, Scott Atran, Douglas Medin, and Khalil Shikaki. 2007. “Sacred Bounds on Rational Resolution of Violent Political Conflict.” Proceedings of the National Academy of Sciences 104 (18): 7357–60. https://doi.org/10.1073/pnas.0701768104.

Hogg, Michael A. 2007. “Uncertainty–Identity Theory.” Advances in Experimental Social Psychology 39: 69–126. https://doi.org/10.1016/S0065-2601(06)39002-8.

Swann, William B., Jr., Ángel Gómez, D. Conor Seyle, J. Francisco Morales, and Carmen Huici. 2009. “Identity Fusion: The Interplay of Personal and Social Identities in Extreme Group Behavior.” Journal of Personality and Social Psychology 96 (5): 995–1011. https://doi.org/10.1037/a0013668.

Tetlock, Philip E., Orie V. Kristel, S. Beth Elson, Melanie C. Green, and Jennifer S. Lerner. 2000. “The Psychology of the Unthinkable: Taboo Trade-Offs, Forbidden Base Rates, and Heretical Counterfactuals.” Journal of Personality and Social Psychology 78 (5): 853–70. https://doi.org/10.1037/0022-3514.78.5.853.

AI Governance, Human Oversight, Metrics, and Contestability

Alon-Barkat, Saar, and Madalina Busuioc. 2023. “Human–AI Interactions in Public Sector Decision Making: ‘Automation Bias’ and ‘Selective Adherence’ to Algorithmic Advice.” Journal of Public Administration Research and Theory 33 (1): 153–69. https://doi.org/10.1093/jopart/muac007.

Autio, Chloe, Reva Schwartz, Jesse Dunietz, Shomik Jain, Martin Stanley, Elham Tabassi, Patrick Hall, and Kamie Roberts. 2024. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile. NIST AI 600-1. Gaithersburg, MD: National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.600-1.

Campbell, Donald T. 1979. “Assessing the Impact of Planned Social Change.” Evaluation and Program Planning 2 (1): 67–90. https://doi.org/10.1016/0149-7189(79)90048-X.

European Parliament and Council of the European Union. 2024. “Regulation (EU) 2024/1689 Laying Down Harmonised Rules on Artificial Intelligence.” Official Journal of the European Union L, July 12, 2024.

Green, Ben. 2022. “The Flaws of Policies Requiring Human Oversight of Government Algorithms.” Computer Law & Security Review 45: 105681. https://doi.org/10.1016/j.clsr.2022.105681.

Manheim, David, and Scott Garrabrant. 2018. “Categorizing Variants of Goodhart’s Law.” arXiv:1803.04585. https://doi.org/10.48550/arXiv.1803.04585.

National Institute of Standards and Technology. 2023. Artificial Intelligence Risk Management Framework (AI RMF 1.0). NIST AI 100-1. Gaithersburg, MD: National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.100-1.

Organisation for Economic Co-operation and Development. 2024. “OECD AI Principles.” Updated May 2024. https://oecd.ai/en/ai-principles.

Parasuraman, Raja, and Dietrich H. Manzey. 2010. “Complacency and Bias in Human Use of Automation: An Attentional Integration.” Human Factors 52 (3): 381–410. https://doi.org/10.1177/0018720810376055.

Skitka, Linda J., Kathleen L. Mosier, and Mark Burdick. 1999. “Does Automation Bias Decision-Making?” International Journal of Human-Computer Studies 51 (5): 991–1006. https://doi.org/10.1006/ijhc.1999.0252.

Strathern, Marilyn. 1997. “‘Improving Ratings’: Audit in the British University System.” European Review 5 (3): 305–21. https://doi.org/10.1002/(SICI)1234-981X(199707)5:3<305::AID-EURO184>3.0.CO;2-4.

Measurement and Research Design

Adcock, Robert, and David Collier. 2001. “Measurement Validity: A Shared Standard for Qualitative and Quantitative Research.” American Political Science Review 95 (3): 529–46. https://doi.org/10.1017/S0003055401003100.

American Educational Research Association, American Psychological Association, and National Council on Measurement in Education. 2014. Standards for Educational and Psychological Testing. Washington, DC: American Educational Research Association.

Campbell, Donald T., and Donald W. Fiske. 1959. “Convergent and Discriminant Validation by the Multitrait-Multimethod Matrix.” Psychological Bulletin 56 (2): 81–105. https://doi.org/10.1037/h0046016.

Krippendorff, Klaus. 2004. “Reliability in Content Analysis: Some Common Misconceptions and Recommendations.” Human Communication Research 30 (3): 411–33. https://doi.org/10.1111/j.1468-2958.2004.tb00738.x.

Messick, Samuel. 1995. “Validity of Psychological Assessment: Validation of Inferences from Persons’ Responses and Performances as Scientific Inquiry into Score Meaning.” American Psychologist 50 (9): 741–49. https://doi.org/10.1037/0003-066X.50.9.741.

Popper, Karl R. 1959. The Logic of Scientific Discovery. London: Hutchinson.

Shadish, William R., Thomas D. Cook, and Donald T. Campbell. 2002. Experimental and Quasi-Experimental Designs for Generalized Causal Inference. Boston: Houghton Mifflin.

Ten-Case Primary Corpus

Altukhov, Yuri P. 2006. Intraspecific Genetic Diversity: Monitoring, Conservation, and Management. Berlin: Springer. https://doi.org/10.1007/3-540-30963-2.

Aumann, Robert J. 1976. “Agreeing to Disagree.” Annals of Statistics 4 (6): 1236–39. https://doi.org/10.1214/aos/1176343654.

Aumann, Robert J. 2005. “War and Peace.” Prize Lecture, Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel, Stockholm, December 8.

Bekenstein, Jacob D. 1973. “Black Holes and Entropy.” Physical Review D 7 (8): 2333–46. https://doi.org/10.1103/PhysRevD.7.2333.

Bekenstein, Jacob D. 1981. “Universal Upper Bound on the Entropy-to-Energy Ratio for Bounded Systems.” Physical Review D 23 (2): 287–98. https://doi.org/10.1103/PhysRevD.23.287.

Bekenstein, Jacob D. 2003. “Information in the Holographic Universe.” Scientific American 289 (2): 58–65.

Bergasa-Caceres, Fernando, Elisha Haas, and Herschel A. Rabitz. 2019. “Nature’s Shortcut to Protein Folding.” Journal of Physical Chemistry B 123 (21): 4463–76. https://doi.org/10.1021/acs.jpcb.8b11634.

Bin-Nun, Yigal. 2016. Kitsur Toldot Yahweh [A Short History of Yahweh]. Tel Aviv: Resling. [Hebrew]. https://www.e-vrit.co.il/Product/7371/%D7%A7%D7%99%D7%A6%D7%95%D7%A8_%D7%AA%D7%95%D7%9C%D7%93%D7%95%D7%AA_%D7%99%D7%94%D7%95%D7%94.

Bin-Nun, Yigal. 2023. Matai Hafakhnu li-Yehudim: Ve-ekh Notsru ha-Datot ha-Monote’istiyot [When Did We Become Jews? And How Did the Monotheistic Religions Emerge?]. Kinneret, Zmora, Dvir. [Hebrew]. https://www.kinbooks.co.il/mti-hpknv-lihvdim.html.

Ehrlich, Avishai. 1992. “The Gulf War and the New World Order.” Socialist Register 28: 227–38. https://socialistregister.com/index.php/srv/article/view/5616.

Ehrlich, Avishai. 2003. “Palestine, Global Politics and Israeli Judaism.” Socialist Register 39: 51–72. https://socialistregister.com/index.php/srv/article/view/5792.

Harsegor, Michael. 1994. Un très petit nombre: Des oligarchies dans l’histoire de l’Occident. Paris: Fayard.

Leibowitz, Yeshayahu. 1992. Judaism, Human Values, and the Jewish State. Edited by Eliezer Goldman. Cambridge, MA: Harvard University Press.

Nobel Prize Outreach. 2005. “The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 2005: Press Release.” October 10.

Orevi, Tomer, Gil Rahamim, Gershon Hazan, Dan Amir, and Elisha Haas. 2013. “The Loop Hypothesis: Contribution of Early Formed Specific Non-Local Interactions to the Determination of Protein Folding Pathways.” Biophysical Reviews 5 (2): 85–98. https://doi.org/10.1007/s12551-013-0113-3.

Tel Aviv University. 2017. “Contemporary Debates in Aesthetics and the Philosophy of Art.” Course syllabus, 2017–2018. https://www30.tau.ac.il/yedion/syllabuse.asp?course=0810600301&year=2017.

Tel Aviv University. n.d. “Michael Harsgor.” Department of General History, In Memoriam. Accessed July 25, 2026. https://humanities.tau.ac.il/history/inmem_harsegor.

Unger, Henry. 1999. “Presentation and Representation in Escher’s Lithographs: The Logic and Aesthetics of Pictorial Nonsense.” Assaph: Studies in Art History 4: 171–90. Publication metadata: https://arts.tau.ac.il/profile/hunger. Full-text scan consulted at https://doczz.net/doc/1601975/dancing-muses%3F.

Zuckermann, Moshe. 1998. Zweierlei Holocaust: Der Holocaust in den politischen Kulturen Israels und Deutschlands. Göttingen: Wallstein.

Zuckermann, Moshe. 2002. Kunst und Publikum: Das Kunstwerk im Zeitalter seiner gesellschaftlichen Hintergehbarkeit. Göttingen: Wallstein.

Zuckermann, Moshe. 2006. “Homogenität und Fragmentierung des Shoah-Gedenkens in Israel.” Zeitgeschichte 33 (2): 99–106.

RATIUM.AI Conceptual Provenance

Dunavich, Benny. 2026a. “A Hidden Split in Formal Reason: Cognitive Duality, Corrective Intelligence, and AI Governance Reliability.” Superseded pre-comparative manuscript. RATIUM.AI internal archive.

Dunavich, Benny. 2026b. “ADM/CIV AI Governance: Civil Corrective Capacity and Soft Closure.” RATIUM.AI. https://www.ratium.ai/articles/adm-civ-ai-governance-civil-corrective-capacity.

Dunavich, Benny. 2026c. “The AI Configuration Paradox.” RATIUM.AI. https://www.ratium.ai/articles/ai-configuration-paradox.

Dunavich, Benny. 2026d. “The Central Equilibrium Problem — Independent Research Framework.” RATIUM.AI. https://www.ratium.ai/articles/cep-doctoral-scale-work.

Dunavich, Benny. 2026e. “The Key to a Stable Governance Layer: Solve the Foundational Decision Problem First.” RATIUM.AI. https://www.ratium.ai/articles/the-key-to-a-stable-governance-layer.

Dunavich, Benny. 2026f. “The Priority of Epistemology: System Convergence and Consensus Ontology.” RATIUM.AI. https://www.ratium.ai/articles/priority-of-epistemology-system-convergence-consensus-ontology.

Dunavich, Benny. 2026g. “When the Correction Mechanism Fails.” RATIUM.AI. https://www.ratium.ai/articles/when-the-correction-mechanism-fails.

Dunavich, Benny. 2026h. “The Sublimation of Ontogenesis.” RATIUM.AI. https://www.ratium.ai/articles/the-sublimation-of-ontogenesis.

Dunavich, Benny. 2026i. “The Ontogenesis Projection Index: Four Canonical Books and the Civilizational Sublimation of Development.” RATIUM.AI. https://www.ratium.ai/foundational-source-dossier/duality-of-innate-cognition/ontogenesis-projection-index.

Related Source and Reference Pages

This article belongs to the public essay layer of RATIUM.AI. For readers who want to move from this article into the broader source, technical, and orientation layers of the project, the following pages provide the relevant entry points.

Articles

The articles page gathers the public essay layer of RATIUM.AI, including arguments on stable AI governance, decision-control architecture, visible governance versus real authority, universal reason, technical competence, purpose governance, and the doctoral-scale framing of CEP.