Every claim, its exact wording, and what it may not say.

EVID-DEC-2024-TIMESTAMP · provenance · prov P3 · scie S0 · repr R2 · orig O0

The 8 December 2024 self-emailed manuscript package

The preserved 8 December 2024 recipient-form message's sender-domain DKIM and Google ARC AMS/Seal mathematics were locally reproduced over their canonicalised signed scopes using keys returned at the signed selectors on 2 August 2026. Attachment identity is supported separately by deterministic MIME decoding and hash crosswalks.

Required qualifications, which travel with this claim

• Its times are the mail servers' own signed record: the sender's DKIM signature at 02:45:30 UTC and Google's ARC seal at 02:45:31 UTC.
• It does not prove the historical DNS state.
• It does not directly authenticate every raw .eml byte, nor independently sign each decoded attachment.
• It does not by itself prove authorship, originality, public disclosure on the signed date, or scientific validity.
• Recipient-side copies have since been acquired and both signing selectors still resolve in DNS; the ARC set bearing d=google.com has no inspected x= expiry and is the preferred provider-key route; sender-signature expiry, historical key-state and live-account custody remain pending.

Wording that may not be used

• Google's live key
• DKIM-verified
• dated, with signature-time assertions that are not an RFC 3161 timestamp (without naming the mechanism)
• tamper-proof

Reviewed 2026-08-02 by the research record · corrections: none

EVID-APR-2025-SUPPLIED · provenance · prov P2 · scie S0 · repr R0 · orig O0

The 30 April 2025 manuscript forensic results

The April sender-DKIM and full Google ARC passes are supplied raw-source forensic results. They were not locally reproduced in the controlling workspace because the raw April recipient-form .eml and executable mutation harness were unavailable there.

Required qualifications, which travel with this claim

• Not locally reproduced. Not independently reproduced.

Wording that may not be used

• April was locally reproduced
• April was independently reproduced

Reviewed 2026-08-02 by the research record · corrections: none

PAPER-IVD-RESULT · empirical · prov P2 · scie S1 · repr R1 · orig O1

Paper IV.d architecture-dependent v4/v5 measurement change

Paper IV.d reports an architecture-dependent v4/v5 measurement change under a bundled protocol: Gemini changed from a significant positive to a significant negative Spearman association, DeepSeek's earlier positive association became null, and GPT remained null. An internal code review in October 2026 found that, for the three models that read flat or negative in the March 2026 run, the scoring endpoint's output cap cut the deepest paraphrases before the judges saw them, and that judges scored paraphrases written by their own model. The early build's laundering-fallback flag is a known false positive (fixed in runner version 5.4.2) and shows nothing about the paraphrase step.

Required qualifications, which travel with this claim

• An internal code review in October 2026 found that, for the three models that read flat or negative in the March 2026 run, the scoring endpoint's output cap cut the deepest paraphrases before the judges saw them, and that judges scored paraphrases written by their own model. The early build's laundering-fallback flag is a known false positive (fixed in runner version 5.4.2) and shows nothing about the paraphrase step.
• Several protocol components changed together.
• The existing comparison does not isolate blinding, laundering or any other individual component.
• Exact-environment reproduction, fresh component ablation and independent replication remain outstanding.

Wording that may not be used

• blinding caused the reversal
• laundering caused the reversal
• two of six models reversed significantly
• successful laundering was demonstrated
• validated benchmark
• independently replicated

Reviewed 2026-10-18 by the research record · corrections: none

PAPER-X-BETA-K · formal · prov P2 · scie S1 · repr R1 · orig O1

Paper X beta > k

Within Eastwood's stated recurrence and minimal model, correction proportional to x^beta asymptotically dominates modelled drift proportional to x^k when beta > k, under the model's assumptions.

Required qualifications, which travel with this claim

• This is a conditional mathematical result, not an empirical finding.
• The exploratory pilots did not validly estimate both exponents. Defects include empty evaluator panels converted to zero in a related v3 run, same-provider OpenAI-family evaluation in another pilot, a mismatch between a coded H1 flag and the stated slope hypothesis, and inadequate capability range.
• RC1's 35/35 tests establish synthetic software behaviour only. They do not make it OSF-ready and do not authorise real collection.

Wording that may not be used

• confirmed mechanism
• universal law
• general alignment proof
• real-model proof
• hard takeoff controlled
• HRIH validated

Reviewed 2026-08-02 by the research record · corrections: none

HRIH-STATUS · interpretive · prov P2 · scie S0 · repr R0 · orig O1

The Hyperspace Recursive Intelligence Hypothesis

HRIH is the author's speculative cosmogenesis hypothesis, labelled speculative in its own text, with an exploratory test agenda.

Required qualifications, which travel with this claim

• Every HRIH presentation must identify its level.
• H0 provenance, H3 analogy and H5 context are never promoted into H4 empirical evidence.
• Use the formal paper's actual public-release date as datePublished. December 2024 is qualified source provenance, not the public paper date.
• Time crystals, quantum error correction, Cauchy forms, scaling analogies, Paper X, religious parallels, BBC material and institutional statements are not HRIH validation.

Wording that may not be used

• HRIH validated
• HRIH confirmed

Reviewed 2026-08-02 by the research record · corrections: none

CONTRIB-BOUNDARY · interpretive · prov P2 · scie S0 · repr R0 · orig O1

The contribution, stated after adversarial prior-work review

Eastwood's claimed contribution is six dated, scoped firsts, each published with the citation that would end it, the strongest by the adversarial review's own verdict being the operationalisation: conjectures converted into runnable benchmarks, a blind-scoring protocol with a published change of direction (no single cause claimed), a bank of drafted registration protocols with kill-conditions frozen ahead of any filing, and public kill-conditions with one recorded as fired; the components are prior work, cited by name, and the claims are to the quantities and the arrangement, not the parts.

Required qualifications, which travel with this claim

• The search is not exhaustive; absence of stronger located prior work is not proof of novelty.
• The alignment and philosophy search is documented; the metabolic-scaling and mathematical-physics search has not been run, and the grey-literature and non-English searches remain incomplete.
• This is academic priority analysis, not a formal legal opinion on originality.

Wording that may not be used

• originator of the embedded-correction alignment thesis
• first preregistered empirical alignment programme
• no prior work
• unprecedented
• uniquely Eastwood

Reviewed 2026-08-02 by the research record · corrections: COR-2026-08-02-003

EQUATION-STATUS · formal · prov P2 · scie S0 · repr R0 · orig O1

U = I x R and U = I x R^alpha

U = I x R^alpha is Law I, the ARC Principle: a form under measurement with an inconclusive exponent estimate.

Required qualifications, which travel with this claim

• Its exponent is still under measurement, and the estimate so far is inconclusive.

Wording that may not be used

• the ARC law
• validated equation
• proven relation
• governs capability

Reviewed 2026-08-02 by the research record · corrections: none

SANDBOX-ISOLATION · method · prov P2 · scie S0 · repr R2 · orig O0

RC1 execution sandbox properties

The RC1 worker provides host confinement through chroot, seccomp, capability dropping and resource limits. Answer-key isolation is a separate property and is being corrected in RC2.

Required qualifications, which travel with this claim

• A candidate can currently recover hidden-test expected outputs from inside the worker using only permitted builtins and attribute access; two of three attack routes score a perfect pass without solving anything.
• The outer isolation defends the host, not the answer key.
• The worker imports libseccomp.so.2 and is Linux-only; it does not run as shipped on macOS and needs the documented disposable container.

Wording that may not be used

• hack-resistant
• tamper-proof
• leak-proof
• cannot be gamed
• prevents test-answer leakage

Reviewed 2026-08-02 by the research record · corrections: none

CONV-TOTALS · method · prov P2 · scie S0 · repr R1 · orig O0

Convergence counting policy

Convergence items are classified before display under schema EVCLASS-2026-08-02. No heterogeneous total is published.

Required qualifications, which travel with this claim

• Any published count is per class and carries its denominator, inclusion rule, dependence grouping, search date and generating commit.

Wording that may not be used

• 22 claimed convergences
• 26 dated events
• 29 convergences
• 19 convergences

Reviewed 2026-08-02 by the research record · corrections: COR-2026-08-02-001 · COR-2026-08-02-002

EVID-RAW-MAILBOX · provenance · prov P3 · scie S0 · repr R2 · orig O0

Raw mailbox object publication policy

No raw mailbox object is published. Each anchor's SHA-256 and byte size are published so a held copy can be checked, and the sealed source is provided to serious verifiers on request.

Required qualifications, which travel with this claim

• A complete mailbox object carries attachments, routing metadata and third-party addresses that cannot be redacted without destroying the byte identity a verifier needs.

Wording that may not be used

• the .eml file is downloadable
• download the .eml

Reviewed 2026-08-02 by operator · corrections: COR-2026-08-02-004

CORPUS-SCALE · factual · prov P3 · scie S1 · repr R2 · orig O0

The size of the published research corpus

Thirty-six documents are counted from the catalogue's data, each in one area and one type: theory (12), empirical work (10), engineering (3), governance (2), overviews and records (6) and other fields (3); twenty-three are papers by their own labels, and thirty-five have a DOI of their own. The count comes from scripts/count-papers.py: every citable work is a document, and a document is a paper when its own label says so. Paper counts and OSF component counts are different quantities.

Required qualifications, which travel with this claim

• Paper counts and OSF component counts are different quantities, counted differently, and are not interchangeable.

Wording that may not be used

• thirty-one research papers
• 31 research papers
• 22 research papers
• sixteen research papers
• eighteen research papers and six supporting frameworks
• eighteen research papers

Reviewed 2026-08-07 by the research record · corrections: none

CONCEPTS-COUNT · factual · prov P2 · scie S0 · repr R1 · orig O1

The number of named concepts introduced across the programme

Thirty-seven named concepts are introduced across the book and the research programme.

Required qualifications, which travel with this claim

• Naming a concept is not the same as originating the mechanism it names.

Reviewed 2026-08-07 by the research record · corrections: none

PREDICTIONS-STATUS · status · prov P2 · scie S0 · repr R0 · orig O0

The status of the twelve tracked predictions

All twelve predictions are withdrawn pending audit. None is currently tracked.

Required qualifications, which travel with this claim

• Withdrawn means withdrawn, not paused. No prediction on this site is currently under active tracking.

Wording that may not be used

• 12 predictions currently tracked
• twelve predictions currently tracked
• predictions currently tracked

Reviewed 2026-08-07 by the research record · corrections: none

PUBLISHING-SCALE · factual · prov P3 · scie S0 · repr R3 · orig O0

The scale of published writing on this site

The site carries 385 blog articles and 623 indexed URLs.

Required qualifications, which travel with this claim

• Counted from the served files and the sitemap on 7 August 2026; both grow.

Wording that may not be used

• 535 indexed URLs
• 1,230 indexed URLs
• 100 blog articles

Reviewed 2026-08-11 by the research record · corrections: none

BOOK-ARTEFACTS · factual · prov P3 · scie S0 · repr R3 · orig O0

The two book artefacts and their lengths

The 30 April 2025 self-emailed manuscript runs to 562 pages. The book published in January 2026 runs to 468 pages.

Required qualifications, which travel with this claim

• The manuscript and the published book are different artefacts with different lengths. Neither number is a correction of the other.

Wording that may not be used

• checkable in any library catalogue

Reviewed 2026-08-07 by the research record · corrections: none

CEILING-DERIVATION-LAW · formal · prov P2 · scie S1 · repr R1 · orig O1

The ceiling of stable self-improvement

The ceiling of stable self-improvement is the reciprocal of the correction shortfall, one minus the correction exponent. Feed the law one assumption, that same-class correctors accumulate corrections the way independent samples do, so the exponent is one half, and it returns two.

Required qualifications, which travel with this claim

• Law first, value second: the reciprocal relation is load-bearing; the digit two is one entailment under one accumulation assumption.
• The asymmetry is deliberate. Kill the value two and the law is untouched: measure the correction exponent and report a different ceiling. Kill the law and the value two has nothing holding it up, because the book printed the square without deriving it and the derivation arrived eighteen months later through the reciprocal. Tests are aimed at the assumption, never at the digit.
• Same-class independence is an assumption, not a result: it is what a fair test must aim at.
• Not an empirical measurement on real correctors. Real-data collection is authorised only when RC2 gates and preregistration approval have both cleared.
• The earlier test 'one number does two jobs, measured two independent ways, and the answers must agree' has been withdrawn as vacuous, in public and before filing. If both exponents arise because accumulation is independence-limited, they share a law and agreement is forced; a test that cannot fail is not a test. The replacement is one accumulation law with two consequences (the conversion exponent and the correction ceiling), which asserts no independence, so the confound has nothing to bite on. The cross-class ratio already on the register carries the unification instead.

Wording that may not be used

• one number does two jobs
• two independent ways
• the answers must agree
• the ceiling equals two
• the critical exponent is two
• derived from the book's R-squared
• measured on real correctors
• law of self-improvement
• runaway

Reviewed 2026-08-12 by the research record · corrections: none