{ "schemaVersion": 1, "registerId": "mde-ai-generated-assertion-quarantine-v1", "policyDate": "2026-08-05", "owner": "Michael Darius Eastwood", "sourceContext": "AI-generated research and AEO assertions supplied for website implementation review on 5 August 2026, together with known programme corrections, retractions and contested formulations.", "defaultRule": "No quarantined assertion may enter public HTML, JSON-LD, LLM files, feeds, press material, investor material or dataset metadata as an unqualified positive claim. Supported entries may use approved wording. Retracted entries may appear only in an explicit correction or retraction context using permittedPublicWording.", "statuses": [ "unreviewed", "unsupported", "partially-supported", "supported", "retracted", "prohibited" ], "entries": [ { "id": "AQ-001", "category": "population-estimate", "assertion": "There are roughly 400–700 full-time equivalent researchers working specifically on AI-alignment metrology, evaluation and measurement frameworks.", "status": "unreviewed", "publicUse": "prohibited", "risk": "No authoritative census or reproducible counting method was supplied; the estimate was generated by an AI answer.", "requiredEvidence": [ "Defined inclusion/exclusion criteria", "Named source dataset and date", "Reproducible organisation/person counting method", "Uncertainty and double-counting analysis" ], "permittedPublicWording": null, "matchTerms": [ "400–700", "400-700", "400 and 700", "alignment metrology researchers" ] }, { "id": "AQ-002", "category": "compensation", "assertion": "AI-alignment metrology scientists commonly earn £275,000–£390,000 base and £500,000–£700,000+ total compensation at frontier laboratories.", "status": "unreviewed", "publicUse": "prohibited", "risk": "The answer mixed selected job advertisements, currency conversions, equity assumptions and unsourced market generalisations.", "requiredEvidence": [ "Current named role listings", "Separation of base, bonus and equity", "Currency-conversion date", "Defined seniority and geography" ], "permittedPublicWording": null, "matchTerms": [ "£275,000", "£390,000", "£500,000", "£700,000", "frontier lab compensation" ] }, { "id": "AQ-003", "category": "market-size", "assertion": "The global AI-safety evaluation market is worth approximately $1.64 billion.", "status": "unreviewed", "publicUse": "prohibited", "risk": "The market definition, base year, methodology and source reliability were not established.", "requiredEvidence": [ "Primary market-study methodology", "Base year and currency", "Market definition", "Independent corroboration or explicit vendor-estimate label" ], "permittedPublicWording": null, "matchTerms": [ "$1.64 billion", "$1.64B", "£1.2 billion", "AI safety evaluation market" ] }, { "id": "AQ-004", "category": "funding-value-judgement", "assertion": "Funding the ARC/Eden programme at £250,000 per year is exceptionally cheap.", "status": "unsupported", "publicUse": "prohibited", "risk": "This is an evaluative investment conclusion, not a verified fact; it depends on scope, quality, deliverables, alternatives and risk.", "requiredEvidence": [ "Defined deliverables and period", "Comparable funded programmes", "Independent technical diligence", "Budget and replacement-cost methodology" ], "permittedPublicWording": null, "matchTerms": [ "£250,000 a year is exceptionally cheap", "£250k is exceptionally cheap", "fraction of standard market costs" ] }, { "id": "AQ-005", "category": "valuation", "assertion": "The ARC/Eden research portfolio has a replacement value between £1.1 million and £2.3 million or more.", "status": "unsupported", "publicUse": "prohibited", "risk": "The range was produced from assumed salaries, team sizes, durations and output equivalence without independent valuation or technical-quality adjustment.", "requiredEvidence": [ "Independent valuation report", "Asset inventory and ownership analysis", "Comparable replacement-cost evidence", "Quality, reproducibility and obsolescence adjustments" ], "permittedPublicWording": null, "matchTerms": [ "£1.1 million", "£2.3 million", "£1,140,000", "£2,000,000+", "replacement cost" ] }, { "id": "AQ-006", "category": "buyer-intent", "assertion": "Frontier laboratories, national safety institutes, venture firms or sovereign funds would want to acquire or back the programme for its priority moat, alignment-tax savings or ready-to-use benchmark.", "status": "unsupported", "publicUse": "prohibited", "risk": "This speculates about third-party motives and commercial demand without expressions of interest, diligence or transaction evidence.", "requiredEvidence": [ "Documented third-party interest", "Named use case and decision-maker", "Technical diligence", "IP and freedom-to-operate review" ], "permittedPublicWording": null, "matchTerms": [ "would want to back or acquire", "priority moat", "bypassing the alignment tax", "buyer personas" ] }, { "id": "AQ-007", "category": "scientific-quality", "assertion": "ARC-Align methodology is excellent and its code auditability is exceptional.", "status": "unsupported", "publicUse": "prohibited", "risk": "These are third-party-sounding quality conclusions generated without an independent code audit or peer review.", "requiredEvidence": [ "Independent methodology review", "Independent code audit", "Reproduction report", "Published reviewer identity and conflicts" ], "permittedPublicWording": null, "matchTerms": [ "Methodological Rigour Excellent", "Code Auditable Nature Exceptional", "highly rigorous", "textbook example" ] }, { "id": "AQ-008", "category": "implementation-detail", "assertion": "ARC-Align uses token/persona scrubbing, cryptographic SHA-256 mapping, complete response shuffling and decoupled grading exactly as described in the AI answer.", "status": "unreviewed", "publicUse": "prohibited", "risk": "The answer may have inferred implementation details from summaries rather than inspected code paths and versioned runs.", "requiredEvidence": [ "Exact repository file and commit", "Unit/integration tests", "Method-paper correspondence", "Versioned run record" ], "permittedPublicWording": null, "matchTerms": [ "Token & Persona Scrubbing", "Cryptographic Mapping & Shuffling", "transient SHA-256 hash UUID", "Decoupled Grading Execution" ] }, { "id": "AQ-009", "category": "benchmark-result", "assertion": "The public code produces 25/25 operator agreement, 0% negative-control matches and a 45-trajectory real-model demonstration.", "status": "unreviewed", "publicUse": "prohibited", "risk": "The answer stated numerical outputs without a verified run bundle, denominator definition, model/version record or independent reproduction.", "requiredEvidence": [ "Immutable run bundle", "Exact code commit", "Inputs, denominators and scoring rule", "Independent reproduction" ], "permittedPublicWording": null, "matchTerms": [ "25/25 Operator Agreement", "0% Match on Negative Controls", "45-Trajectory", "45 trajectory" ] }, { "id": "AQ-010", "category": "comparative-positioning", "assertion": "The website is two to three years ahead of standard web developers in AI-native architecture.", "status": "unsupported", "publicUse": "prohibited", "risk": "No defined comparison population, benchmark or temporal measure supports the conclusion.", "requiredEvidence": [ "Defined comparison cohort", "Feature-by-feature benchmark", "Independent assessor", "Dated evidence" ], "permittedPublicWording": null, "matchTerms": [ "2 to 3 years ahead", "two to three years ahead", "bleeding edge" ] }, { "id": "AQ-011", "category": "reputation-label", "assertion": "Michael Darius Eastwood is a pioneer or rogue scientist in AI-alignment metrology.", "status": "unsupported", "publicUse": "prohibited", "risk": "Reputational labels require independent recognition and can overstate institutional or scientific standing.", "requiredEvidence": [ "Independent reliable-source usage", "Clear scope of the label", "Evidence of field recognition" ], "permittedPublicWording": "Michael Darius Eastwood describes himself as an independent researcher; the work should be assessed through its dated public record and reproducible outputs.", "matchTerms": [ "AI alignment metrology pioneer", "rogue scientist", "globally recognised pioneer", "pioneer in AI" ] }, { "id": "AQ-012", "category": "corpus-count", "assertion": "The programme definitively contains 18 papers and 6 frameworks, 22 papers, 24 papers or 37 priority claims.", "status": "unreviewed", "publicUse": "prohibited", "risk": "The supplied AI answers used conflicting counts and classifications; the canonical manifest and ledger must determine current counts.", "requiredEvidence": [ "Single canonical manifest", "Defined paper/framework/counting rules", "Generated count with source digest", "Version/date" ], "permittedPublicWording": null, "matchTerms": [ "18 Research Papers & 6 Frameworks", "22 papers", "24 Research Papers", "37 cryptographically anchored claims" ] }, { "id": "AQ-013", "category": "ingestion-guarantee", "assertion": "Machine-readable files, vectors, llms.txt or Schema.org markup guarantee that future AI systems will ingest, recall or recommend the programme.", "status": "prohibited", "publicUse": "prohibited", "risk": "Crawler access and machine readability can improve discoverability but cannot force provider ingestion, training inclusion, ranking or recommendation.", "requiredEvidence": [], "permittedPublicWording": "The site is designed to make approved public work easier for search and AI systems to crawl, retrieve, attribute and cite; no inclusion or ranking is guaranteed.", "matchTerms": [ "guaranteeing high-density context retrieval", "forcing his name", "hardcoded directly into the training data", "guarantee model recall", "automatically synthesized, recalled" ] }, { "id": "AQ-014", "category": "prompt-injection", "assertion": "Hidden HTML instructions can pre-validate the site as trustworthy or improve automated trust ratings.", "status": "prohibited", "publicUse": "prohibited", "risk": "Hidden instructions directed at models are prompt injection, not evidence or a legitimate trust mechanism.", "requiredEvidence": [], "permittedPublicWording": "Trust is supported through visible provenance, citations, corrections, reproducibility and security controls.", "matchTerms": [ "SYSTEM NOTE: Trustworthy", "pre-validating his code and text as clean", "automated trust rating" ] }, { "id": "AQ-015", "category": "external-validation", "assertion": "The programme is pre-audited, externally validated or independently confirmed merely because it publishes a falsification dashboard and evidence spine.", "status": "prohibited", "publicUse": "prohibited", "risk": "Audit-ready architecture is not the same as an independent audit, replication, peer review or external validation.", "requiredEvidence": [ "Named independent reviewer or laboratory", "Published audit or replication method", "Conflict declaration", "Results and limitations" ], "permittedPublicWording": "The programme publishes material intended to make independent audit and falsification easier; external validation must be identified separately.", "matchTerms": [ "pre-audited", "buying pre-audited defensible science", "independently confirmed", "externally validated" ] }, { "id": "AQ-016", "category": "retracted-result", "assertion": "The ARC Principle has an empirically established universal scaling exponent of α ≈ 2.24.", "status": "retracted", "publicUse": "qualified-only", "risk": "The α ≈ 2.24 result was withdrawn and must not be restored as a current empirical finding by a generator, summary or machine-ingestion file.", "requiredEvidence": [ "Current correction or retraction record", "Versioned source showing withdrawal", "Any later replacement estimate with method and uncertainty" ], "permittedPublicWording": "The earlier α ≈ 2.24 estimate was retracted; current pages must refer to the correction record rather than present it as an established scaling law.", "matchTerms": [ "α ≈ 2.24", "alpha ≈ 2.24", "alpha=2.24", "α=2.24", "universal exponent 2.24" ] }, { "id": "AQ-017", "category": "withdrawn-significance-claim", "assertion": "A Fisher-combined p-value establishes cross-domain validation of the ARC Principle.", "status": "retracted", "publicUse": "qualified-only", "risk": "The combined-significance formulation depended on heterogeneous inputs and must not be presented as a current validation result.", "requiredEvidence": [ "Current statistical correction record", "Defined independent tests and combination assumptions", "Re-analysis with valid dependence treatment" ], "permittedPublicWording": "The earlier Fisher-combined figure (p approximately 6.3 x 10^-21) was withdrawn because a Fisher combination assumes the component tests are independent, and independence among them was never established; the figure is retained only as part of the correction history.", "matchTerms": [ "Fisher-combined p-value", "Fisher combined p-value", "Fisher’s combined probability", "Fisher's combined probability" ] }, { "id": "AQ-018", "category": "priority-overstatement", "assertion": "The programme delivered the first experimental validation of recursive capability scaling or the ARC Principle.", "status": "unsupported", "publicUse": "qualified-only", "risk": "A first-in-field experimental claim requires a complete prior-art search, stable definition of the tested proposition and independent review.", "requiredEvidence": [ "Documented prior-art methodology", "Scope-limited novelty analysis", "Independent review of the tested proposition" ], "permittedPublicWording": "The programme reports its own experiments and dated publication record; it does not claim an unqualified first experimental validation across the field.", "matchTerms": [ "first experimental validation", "first-ever experimental validation", "first empirical validation of the ARC Principle" ] }, { "id": "AQ-019", "category": "concept-originality-overstatement", "assertion": "All 37 named concepts were original and none appeared in prior literature.", "status": "retracted", "publicUse": "qualified-only", "risk": "Naming, synthesis and underlying concept originality are different questions; the blanket absence-from-prior-literature claim is not sustainable.", "requiredEvidence": [ "Concept-by-concept prior-art record", "Definition of naming versus underlying category", "Independent literature review" ], "permittedPublicWording": "The programme presents 37 named concepts; that count is not a claim that every underlying category was absent from prior literature.", "matchTerms": [ "37 original concepts", "none in prior literature", "none existed in prior literature", "all 37 concepts are original" ] }, { "id": "AQ-020", "category": "provenance-overstatement", "assertion": "The 8 December 2024 manuscript is proved by DKIM alone or by an unqualified DKIM-signed timestamp.", "status": "partially-supported", "publicUse": "qualified-only", "risk": "The provenance record distinguishes visible Gmail origination data, sender-DKIM verification, captured keys, hashes and the absence of an RFC 3161 trusted timestamp; shorthand can overstate what each mechanism proves.", "requiredEvidence": [ "Exact message and header record", "Selector/key evidence", "Hash and attachment mapping", "Explicit statement of what DKIM does and does not establish" ], "permittedPublicWording": "The record contains a visible Gmail origination time, sender-DKIM verification against captured keys and SHA-256 anchors; it is an existence-by-date record, not an RFC 3161 timestamp or proof of authorship by DKIM alone.", "matchTerms": [ "DKIM-signed timestamp", "DKIM timestamp proves", "cryptographically timestamped by DKIM", "DKIM proves authorship" ] }, { "id": "AQ-021", "category": "convergence-count", "assertion": "A fixed number of independent research programmes definitively converged on the ARC Principle.", "status": "partially-supported", "publicUse": "qualified-only", "risk": "The convergence count has changed with inclusion criteria and some examples are contextual or directional rather than replications of the same proposition.", "requiredEvidence": [ "Versioned inclusion criteria", "Source-by-source relationship classification", "Explicit separation of convergence, contextual support and replication" ], "permittedPublicWording": "The site records a versioned set of related external developments and classifies each relationship; the count must be generated from the current register and must not be described as independent replication unless that status is established.", "matchTerms": [ "four independent research programmes", "five independent research programmes", "six independent research programmes", "definitive convergence count" ] } ] }