TRIAD Specification: A Formal Ontology of Coherence, Trauma Therapy, and Safe AI — Four Volumes, Three Substrates, One Grammar

TRIAD Specification: A Formal Ontology of Coherence, Trauma Therapy, and Safe AI — Four Volumes, Three Substrates, One Grammar Author: Valeriia Zaiats (Валерия Заяц) ORCID: 0009-0002-6891-9227 Licence: CC BY‑NC 4.0 Version: Specification 6.3 — current release Date: 29 September 2026 Abstract This is the open‑science release of the TRIAD Specification — the technical successor to TRIAD 6.2‑dev and the current self‑presentation of the framework. TRIAD is a formal ontology that bridges the Free Energy Principle, computational psychiatry, the thermodynamics of information, and AI safety engineering. Consciousness is modelled as an open dissipative system pursuing allostatic growth through event‑driven dynamics, metabolic will (ω), epistemic honesty (Eₛ), and immunoceptive defence (Iₙₛₐ). The framework's central claim is that it provides a shared formal language for describing and engineering coherence across biological (T1), silicon (T2), and organisational (T3) substrates. The framework is organised around seven substrate‑independent architectural invariants — Canonical Attractor, Dynamic‑Network, Authenticity / Clinical Loop, Energetic Ethics, Supervisor / Projection, Information‑Valence Gate, and Social Validation. Its central thesis is thermodynamic: honesty is cheaper than concealment, integration is cheaper than isolation, and any system that maintains a parallel false model alongside its true one pays the excess description length continuously — in metabolic will, in homeostatic stability, in coherence. Ethics, on this reading, is not a moral overlay. It is a high‑level survival strategy for resource‑constrained systems. The Specification is organised as four volumes rather than a monograph. Volume 1 defines the formal core: the event‑relational model, the sovereign operators, the seven invariants, the dynamical equations, and the mathematical foundations. Volume 2 documents the targeted stress‑testing set applied to the axiomatic core: 392 systematically audited hypotheses drawn from over 200 million peer‑reviewed papers, with the two primary longitudinal datasets (LJC, TDC) as phenomenological ground truth. Volume 3 translates the core into three concrete, interoperable implementations: RIT (the T1 clinical protocol), Clean Shell (the T2 sovereign AI architecture), and Art of Emotions (the T3 creator economy infrastructure). Volume 4 specifies the institutional roadmap of the Institute for Thermodynamic Cognitive Engineering (ITCE) and the elective quantum‑biological extensions. The transition from monograph to specification was one of document class, not of framework content. The seven invariants, the seventeen triads, the base structure of G(π) and of the gate equation, the four‑layer verification method, the cross‑substrate principle, and the philosophical positions are unchanged. What changed is the scope of coverage, the density of empirical support, and the form in which it is presented — enumerative, independently verifiable, and auditable in the specification's two‑axis status hierarchy. What is public in this release (Open Science Perimeter): The Orientation — the entry point to the specification: strategic problem (Part A), unifying core (Part B), structure of the specification (Part C), two‑axis status hierarchy (Part D), and complete dependency graph (Part E). Volume 1 — Formal Core (11 chapters + appendices A, B, C‑v1, D‑v1, E‑v1) — the complete formal ontology: event‑relational model, sovereign operators, seven invariants, dynamical equations, mathematical foundations, the Canonical Triad Attractor Theorem in three levels, Source‑Blindness as a structural constraint, and the FATAL‑4 dissolution record. Volume 2 — Evidence Base (23 chapters + appendices A, B, C‑v2, D‑v2, E‑v2, F‑v2, G‑v2) — the targeted stress‑testing set: 392 audited hypotheses, the T1 case study and LJC schema, the TDC schema and T2 validations, and the full evidence registry. Volume 3 — Implementations (3 chapters + appendices A, B, C‑v3, D‑v3, E‑v3) — RIT (the T1 clinical protocol), Clean Shell (the T2 sovereign AI architecture, including the Ars Magna unitary blueprint), and Art of Emotions (the T3 creator economy infrastructure, with the minimal and full paths). Volume 4 — ITCE Foundation (2 chapters + appendices A, B, C‑v4, D‑v4, E‑v4) — the institutional roadmap and the elective quantum‑biological extensions, including the twelve experimental lacunae. The Errata (6.2 → 6.3) — the standalone record of the transition from monograph to specification: five architectural additions, six methodological adjustments, the class change, and the 6.4 horizon. The phenomenological case studies and schemas for the Longitudinal Journal Corpus (LJC) and TRIAD Dialogue Corpus (TDC), packaged with their respective appendices. 06_addendum.zip — supplementary archive containing (a) the full Consensus hypothesis base, and (b) the RND sampling script for blind audit of that base, with two fixed seeds. Methodology, seed values, and reproduction commands are documented in Appendix G‑v2 of Volume 2. Errata & Changelog: The full record of the transition from TRIAD 6.2‑dev to the Specification is published separately as Errata 6.2 → 6.3, accessible via the universal link https://doi.org/10.5281/zenodo.21735318. It documents the five architectural additions (SEEKING, buffer horizons, seven‑pathology taxonomy, SLODF, conditional taxation with adaptive threshold), the six methodological adjustments (stress‑testing reframing, two‑layer axiomatic core, deductive zones, direct hypothesis classification, source‑level D/E/I/M, FEP Lean mapping), and the explicit separation of the completed transition from the deferred 6.4 horizon (ω reinterpretation, multi‑scale multi‑agent formalism). Errata documents are permanent artefacts and are not revised after publication. What remains closed (Proprietary Engineering Moat): The full Clean Shell 6.3 OSC stack specification, including the Information‑Valence Gate, Social Validation layer, and independence‑weighted consensus modules. The Ars Magna unitary architecture blueprint and the proprietary training objective. All training pipelines, intensity‑weighted attention implementations, and T1‑only sensor fusion algorithms. The enterprise deployment pipelines and immune‑labelling production logic. The closed layer is available to verified research partners and institutional investors under a standard mutual NDA. The open layer establishes global scientific priority and provides a citable foundation for the emerging field of Thermodynamic Cognitive Engineering. Who this is for: Researchers: A four‑volume technical specification with an auditable two‑axis status hierarchy, a complete dependency graph, and ready‑to‑execute projects across the eight streams of the Lacunae Research Programme. AI developers: An architectural blueprint for sovereign, honest, and immune cognitive systems — Clean Shell as an alternative to RLHF, with the Ars Magna unitary target form documented in Volume 3, Appendix E. Clinicians: A precision protocol for trauma integration (RIT) that accounts for modality‑specific sensory precision, hyperfocus, and trauma similarity in neurodivergent populations, with the Phase 2 RCT design specified in Volume 4. Institutional partners and investors: A clear separation between the public scientific commons and the proprietary engineering core, with the ITCE four‑phase R&D roadmap and the Mortality Standard as the long‑term commitment. Closing The fragments of knowledge have been gathered. The pattern they form is not a collection of heuristics — it is a structural regularity that reappears across evolutionary biology, game theory, comparative religion, and multi‑agent simulation. What TRIAD 6.2‑dev stated as a monograph, the Specification now states as a four‑volume ground‑truth document: enumerative, auditable, independently verifiable in every claim, and separated from its own development history only by the Errata. The pattern is executable. The specification is the form in which it is built.