--- # TRUST.md — epistemic trust declaration for the Neuronautix knowledge base. # Convention: TRUST.md v0.4.0-rc.1 — https://github.com/Neuronautix/TRUST.md # Canonical repository filename is TRUST.md; canonical served path is /trust.md. # # Migrated from this repository's v0.1 declaration on 2026-08-02. The v0.1 # source is preserved unchanged in git history (see x_migration below). # # v0.4 separates the EVIDENCE SUBJECT from ASSESSMENTS about that subject. # This declaration registers its subject and publishes ZERO assessments: the # facts a conforming v0.4 assessment requires — a versioned protocol # identifier, inspectable evidence-basis links, and an explicit independence # declaration — do not yet exist for this corpus, and the specification # forbids inventing them. An empty assessment collection means no assessment # has been made. It does not mean the evidence was assessed as low quality. trust_md_version: "0.4" title: "Neuronautix — Trust & Epistemic Provenance Declaration" description: > How knowledge and inferences in this repository are produced, categorised, and reviewed. Covers the source-backed knowledge base, the cited analytical notes and their inline per-claim epistemic markup, and the presentation packages. canonical: "https://neuronautix.com/trust.md" license: "Apache-2.0" companions: fair: "/fair.md" ai_declaration: "/AI-DECLARATION.md" inline_markup_spec: "/notes/EPISTEMIC-MARKUP.md" # ── Who and what produces the knowledge (provenance of authorship) ── produced_by: humans: - name: "Damien Huzard, PhD" role: "author, domain reviewer, accountable signatory" orcid: "0000-0003-4820-7951" org: "Neuronautix" agents: - name: "Claude (Anthropic), via Claude Code" role: "retrieval, drafting, epistemic markup, summarisation" oversight: "human-reviewed" # ── Governance ── governance: source_of_truth: > The human-approved record — not raw model output — is the source of truth. AI-assisted text is retrieval-grounded and verified before publication. Authorship of the software and site itself is declared separately in /AI-DECLARATION.md. no_fabricated_citations: true review_policy: > Every published note passes two human-in-the-loop gates — an outline gate and a draft gate — defined in .github/instructions/note-review-agent. instructions.md. The reviewing agent produces findings and cannot approve; only the accountable human can clear a gate. Citation integrity is checked mechanically in CI by scripts/check_note_citations.py. correction_policy: > Errors are corrected in place with the change recorded in git history. Substantive corrections are dated. This declaration is versioned with the repository; superseded declarations remain retrievable in git history rather than being overwritten silently. conflict_of_interest: > Neuronautix is a scientific consultancy and publishes two of the three manifest conventions referenced here (trust.md and fair.md). Promotional or normative statements are labelled with the 'view' category rather than presented as cited fact. # ── The epistemic model (mirrors the inline Trust Lens markup) ── epistemic_model: categories: - {id: "cited", label: "Cited fact", definition: "Directly supported by a cited source."} - {id: "consensus", label: "Established consensus", definition: "Widely accepted domain knowledge or a standard definition."} - {id: "inference", label: "Inference", definition: "Reasoned from one or more sources; not stated verbatim."} - {id: "hypothesis", label: "Hypothesis", definition: "A forward-looking or speculative claim."} - {id: "view", label: "Neuronautix view", definition: "An explicit interpretation, position, or normative conclusion."} support_bands: - {id: "speculative", label: "Speculative", meaning: "Normative, visionary, or opinion claim with little direct evidence.", range: "0-29"} - {id: "tentative", label: "Tentative", meaning: "Plausible forward-looking claim with partial support.", range: "30-49"} - {id: "moderate", label: "Moderate", meaning: "Reasonable inference, or consensus without a pinpoint citation.", range: "50-69"} - {id: "high", label: "High", meaning: "Stated in a cited source, secondary or lightly interpreted.", range: "70-89"} - {id: "very-high", label: "Very high", meaning: "Directly stated in a primary, peer-reviewed, or regulatory source.", range: "90-100"} confidence_scale: type: "integer" range: [0, 100] independent_of_category: true not_probability: true encoding: inline: 'HTML spans: ' reader_tooling: "Trust Lens (assets/js/trust-lens.js) — toggleable highlight, colour-blind-safe palette, per-note summary" alignment: - "W3C PROV / PAV — provenance and authoring of assertions" - "Nanopublications — assertion + provenance + publication info" - "SEPIO (Scientific Evidence & Provenance Information Ontology)" - "Evidence & Conclusion Ontology (ECO)" - "schema.org ClaimReview — emitted per note as JSON-LD" # ── The evidence subject this declaration is about ── subjects: - id: "neuronautix-knowledge-base" type: "repository" identifier: "https://github.com/Neuronautix/neuronautix.com" version: "d939d3a1ddc1461779205df02cd231716e833423" # ── No assessment is published. See the header note and limitations. ── assessments: [] limitations: - "No conforming v0.4 assessment is published: this declaration identifies its subject and makes no assessment claim about it." - "The per-claim categories and confidence values summarised in x_inline_markup_profile are author/reviewer judgements recorded in the page markup. They are descriptive metadata, not a v0.4 assessment, and carry no protocol, evidence basis, or independence declaration." - "A single 0-100 value conflates evidence strength and stated confidence. v0.4's ordinal support bands are the primary axis; the integer is a non-probabilistic refinement only." - "Self-declared and not independently audited. Conformance to TRUST.md is not certification of accuracy, methodological quality, independence, or reproducibility." last_reviewed: "2026-08-02" # ── Extensions (x_ namespace; not part of the v0.4 assessment contract) ── # Provenance of this declaration's own migration. x_migration: from_version: "0.1" to_version: "0.4" migrated_on: "2026-08-02" source_declaration: "https://github.com/Neuronautix/neuronautix.com/blob/d939d3a1ddc1461779205df02cd231716e833423/trust.md" notes: > The v0.1 declaration carried a single corpus.average_trust aggregate, deprecated at v0.3. It is not mapped to a v0.4 assessment aggregate. The per-artifact profiles are retained below as descriptive extension data and were recomputed from the published markup at migration time; the v0.1 file reported 19 notes and 440 claims, which had become stale. # Descriptive summary of the inline per-claim markup, recomputed 2026-08-02 # directly from the published note HTML. Not an assessment. x_inline_markup_profile: method: "Counted from data-epi and data-trust attributes in notes/*/index.html, excluding reference lists." computed_on: "2026-08-02" notes_with_markup: 31 total_claims: 764 mean_confidence: 73 category_distribution: {cited: 379, view: 213, inference: 131, consensus: 22, hypothesis: 19} per_artifact: - {path: "/notes/2026-05-namo-new-approach-methodology-ontology/", claims: 40, avg: 88, dist: {cited: 36, inference: 3, view: 1}} - {path: "/notes/2026-08-ai-transparency-declarations/", claims: 24, avg: 84, dist: {cited: 18, view: 6}} - {path: "/notes/2026-05-nam-data-management-ind-submissions/", claims: 29, avg: 83, dist: {cited: 22, inference: 4, view: 3}} - {path: "/notes/2026-06-trust-md-manifest/", claims: 28, avg: 82, dist: {cited: 15, consensus: 1, inference: 2, view: 10}} - {path: "/notes/2026-05-llm-kg-metadata-architecture/", claims: 43, avg: 81, dist: {cited: 38, consensus: 1, inference: 2, view: 2}} - {path: "/notes/2026-06-data-welfare-animal-welfare/", claims: 25, avg: 80, dist: {cited: 15, consensus: 2, hypothesis: 1, inference: 6, view: 1}} - {path: "/notes/2026-06-dvc-metrics-explained/", claims: 22, avg: 80, dist: {cited: 11, consensus: 2, inference: 6, view: 3}} - {path: "/notes/2026-05-after-the-knowledge-graph/", claims: 46, avg: 78, dist: {cited: 21, consensus: 5, inference: 5, view: 15}} - {path: "/notes/2026-06-ai-home-cage-monitoring/", claims: 14, avg: 78, dist: {cited: 11, view: 3}} - {path: "/notes/2026-05-neuronautix-as-research-infrastructure/", claims: 23, avg: 77, dist: {cited: 15, inference: 3, view: 5}} - {path: "/notes/2026-06-dvc-analysis-pipelines/", claims: 33, avg: 77, dist: {cited: 19, inference: 3, view: 11}} - {path: "/notes/2026-06-fair-md-manifest/", claims: 27, avg: 77, dist: {cited: 8, consensus: 1, hypothesis: 1, inference: 3, view: 14}} - {path: "/notes/2026-05-ai-ready-nam-data/", claims: 20, avg: 76, dist: {cited: 10, inference: 5, view: 5}} - {path: "/notes/2026-05-meet-the-founder-damien-huzard/", claims: 15, avg: 76, dist: {cited: 12, view: 3}} - {path: "/notes/2026-06-fair-trust-md-manifests/", claims: 28, avg: 75, dist: {cited: 10, consensus: 1, hypothesis: 1, inference: 4, view: 12}} - {path: "/notes/2026-05-fairscape-biocompute-nam-provenance/", claims: 15, avg: 74, dist: {cited: 4, inference: 6, view: 5}} - {path: "/notes/2026-05-metadata-as-a-nam/", claims: 35, avg: 74, dist: {cited: 14, consensus: 2, inference: 9, view: 10}} - {path: "/notes/2026-05-data-first-fair-by-design/", claims: 12, avg: 72, dist: {cited: 5, inference: 2, view: 5}} - {path: "/notes/2026-05-metadata-as-nam/", claims: 11, avg: 72, dist: {cited: 2, consensus: 3, inference: 3, view: 3}} - {path: "/notes/2026-05-schema-first-metadata-agent-nam/", claims: 11, avg: 72, dist: {cited: 4, inference: 4, view: 3}} - {path: "/notes/2026-07-arrive-vs-prepare/", claims: 18, avg: 72, dist: {cited: 7, consensus: 1, hypothesis: 1, inference: 4, view: 5}} - {path: "/notes/2026-05-post-hoc-curation-fails-nam-platforms/", claims: 15, avg: 70, dist: {cited: 6, inference: 5, view: 4}} - {path: "/notes/2026-05-linkml-preclinical-metadata/", claims: 8, avg: 68, dist: {cited: 4, inference: 2, view: 2}} - {path: "/notes/2026-05-nam-metadata-mandatory-qualification/", claims: 21, avg: 68, dist: {cited: 7, inference: 10, view: 4}} - {path: "/notes/2026-07-unusable-animal-data-welfare/", claims: 23, avg: 68, dist: {cited: 14, consensus: 1, inference: 3, view: 5}} - {path: "/notes/2026-07-ethical-debt-book-preview/", claims: 31, avg: 63, dist: {cited: 13, hypothesis: 1, inference: 5, view: 12}} - {path: "/notes/2026-06-wellfair-webinar-recap/", claims: 12, avg: 62, dist: {cited: 3, hypothesis: 1, view: 8}} - {path: "/notes/2026-06-ethical-debt-preclinical-research/", claims: 64, avg: 60, dist: {cited: 17, consensus: 2, hypothesis: 2, inference: 16, view: 27}} - {path: "/notes/2026-05-human-loop-ai-ready-preclinical-data/", claims: 11, avg: 59, dist: {cited: 4, inference: 3, view: 4}} - {path: "/notes/2026-05-nam-evidence-commons-tool-stack/", claims: 34, avg: 55, dist: {cited: 7, hypothesis: 1, inference: 10, view: 16}} - {path: "/notes/2026-06-hcm-2050-vision/", claims: 26, avg: 54, dist: {cited: 7, hypothesis: 10, inference: 3, view: 6}} --- # trust.md — Neuronautix Trust & Epistemic Provenance This file declares **how the knowledge in this repository is produced, categorised, and reviewed** — and what has *not* been assessed. It is the repository-level companion to the inline *Trust Lens*, the per-claim epistemic markup readers can toggle on any note. The YAML block above is machine-readable; this prose is for people. It follows the **[TRUST.md convention](https://github.com/Neuronautix/TRUST.md), v0.4.0-rc.1** — a release candidate, and an experimental convention. This repository was migrated from the v0.1 declaration on 2 August 2026. ## What changed in the migration, and why it matters The v0.1 declaration published a single corpus-wide average confidence score. The convention has since moved away from exactly that: v0.3 demoted the 0–100 integer to an optional, explicitly non-probabilistic refinement inside an ordinal band and deprecated `corpus.average_trust`; **v0.4 separates the evidence subject from assessments about it**, allows independent or conflicting assessments to coexist without averaging, and forbids impact metrics — citations, downloads, reuse — from determining quality or fitness. The reason is worth stating plainly: **evidence does not have one universal level of trust.** The same record can be fit for one purpose and unfit for another, and a single headline number hides that rather than reporting it. ## This declaration publishes no assessment A conforming v0.4 assessment requires facts that do not yet exist for this corpus: a versioned protocol identifier, inspectable evidence-basis links, an explicit independence declaration, and — for any fitness claim — a declared purpose. The specification forbids inventing them, so none were invented. `assessments: []` therefore means **no assessment has been published**, not that the evidence was assessed and found wanting. Registering the subject honestly and leaving the assessment collection empty is an explicitly valid starting point under v0.4. ## What is still published: the inline markup profile Every substantive claim in a note carries two independent attributes in the page markup: 1. **Category** — `cited` · `consensus` · `inference` · `hypothesis` · `view`. 2. **Confidence** — an integer 0–100, bucketed into the five ordinal support bands. As recomputed on 2 August 2026 directly from the published pages: **31 notes · 764 graded claims · mean confidence 73**, distributed 50% cited · 28% Neuronautix view · 17% inference · 3% consensus · 2% hypothesis. The per-note table in the YAML above makes this auditable page by page. That data sits under an `x_` extension key deliberately. It is descriptive metadata about how the author marked up their own prose. It carries no protocol, no evidence basis, and no independence declaration, so under v0.4 it is not an assessment and is not presented as one. The most evidence-dense notes average in the high 80s; the explicitly forward-looking ones score lowest **by design**. A low mean here is a feature — it tells the reader the piece is a position or a prediction rather than a literature report. ## Relationship to the companion manifests - **[`/AI-DECLARATION.md`](/AI-DECLARATION.md)** — *who wrote the software and the site*, at what level of autonomy ([AI-DECLARATION.md](https://ai-declaration.md/en/0.1.2/) v0.1.2). - **`trust.md`** (this file) — *what epistemic status the published knowledge has*. - **[`/fair.md`](/fair.md)** — *whether the data and knowledge can be found and reused* (fair.md v0.3.1). ## Relationship to existing standards TRUST.md is a lightweight front door, not a competitor, to the formal assertion/provenance stack: **W3C PROV / PAV** for provenance of assertions; **nanopublications** for the assertion + provenance + publication-info pattern; **SEPIO** and the **Evidence & Conclusion Ontology** for evidence typing; and **schema.org ClaimReview**, which this site emits as JSON-LD on every note. It is deliberately distinct from JournalList's **`trust.txt`**, which declares an organisation's trusted *relationships* rather than the epistemic status of its content. ## Conformance is not certification A valid TRUST.md file does not establish that the evidence it identifies is true, methodologically sound, independent, reproducible, or fit for any undeclared use. This declaration is self-reported and has not been independently audited. ## Changelog - **2026-08-02** — migrated v0.1 → v0.4.0-rc.1. Subject registry added; assessment collection left empty; deprecated `corpus.average_trust` retired; per-claim profile recomputed from the published pages (31 notes / 764 claims, replacing a stale 19 / 440) and moved to an `x_` extension; `/AI-DECLARATION.md` added as a companion. - **2026-06-06** — v0.1 declaration published as the reference implementation of the then-current convention.