The funders brief

What this programme is, in one page

The ARC and Eden Research Programme is a single-investigator, evidence-gated programme testing one idea: that capability in recursive systems scales with recursive depth, and that alignment must therefore be embedded in the loop rather than imposed from outside. Every claim is dated, hashed and publicly anchored. Results publish with kill conditions fixed in advance, corrections are logged in public, and the whole record invites independent replication.

What you would be funding is not a theory. It is the instrument that would test one. Every located predecessor stops at a requirement, a proposal, a defence method or a derivation. The quantities are stated here, on a dated record, each with the test that would refute it: d/(d+1) argued as a mathematical necessity rather than a property of any mechanism, β > k as a stability condition that can be measured on a running system, and α ≤ 2 as the numerical ceiling that condition implies at the limit of how well any corrector inside the system can scale.

One honest boundary travels with the criterion: it certifies that the brakes keep pace with the engine, never that the driver steers to the right place. It is a correctability measure. The correction target carries its own separate burden of proof.

Why this deserves money before anything else does: no located predecessor computes a stability criterion for self-improving systems from the corrector’s own measured scaling exponent, the thing that decides whether every other safeguard means anything. This programme has one on the table, prepared for registered measurement under ARC-Beta-k, its measurement arm.

What is claimed, at full strength: a conversion law for realised capability, U = I × R^α, capability as intelligence converted through recursion; a stability ceiling that is the reciprocal of the corrector’s shortfall from full proportionality, which, on the same-class independence assumption, resolves to two; and one accumulation law with two consequences, the conversion exponent and the correction ceiling, replacing an earlier collapsed-identity framing withdrawn as vacuous before filing: if both exponents arise because accumulation is independence-limited, they share a law, agreement is forced, and a test that cannot fail is not a test. The full withdrawal history lives on the whole-story report. If the registered measurements confirm it, it is the first stability criterion for self-improving systems computed from the corrector’s own measured scaling exponent, and the conversion law’s first measured exponent.

The synthesis is the claim. Components have prior work; the arrangement is the contribution. The related work page sets out, line by line, where each component was already stated, and where the joint pattern and the scaling relation were not.

Sources: research/for-funders.html · research/index.html

The record so far

A dated ledger of what the programme has produced with no external funding. Every row carries its check.

8 December 2024. A five-file manuscript package, ARC-sealed and OpenTimestamps-anchored. 221,236 words in aggregate across the package, carrying the embedded-correction alignment thesis, the ARC Hypothesis by name, and the control-failure passage. OpenTimestamps receipts anchored 6 August 2026. Sent-side SHA-256 begins f0d1f38f.

30 April 2025. The book manuscript, dated in transit. The full draft of Infinite Architects crossed two providers into Gmail, collecting an independently signed authentication verdict on the way.

2 January 2026. Infinite Architects, published. 468 pages. Thirty-seven named concepts, each with a dated record in the priority ledger. The author's printed prior statement of the exponent, never described as a preregistration. ISBN 978-1-80605-620-0. Written at night, between court deadlines.

January to July 2026. Nineteen papers deposited under the programme DOI, each with its own. Fourteen in a numbered sequence, two lettered, three supporting. A principle, a test of it, what broke, and what survived the breaking. Every format that exists is served free. Every paper states the condition that would kill it.

March 2026. The strongest pilot: fifty domains against one prediction. The composition operator of each system classified before fitting, the scaling family predicted from it, then tested by independent model fitting. Nineteen of twenty-five empirical domains matched under strict AICc, permutation p = 7 × 10⁻⁵. The misses are analysed in the paper rather than hidden.

24 March 2026. The convergence argument was sent to Professor Lloyd Demetrius at Harvard and Professor Geoffrey West at the Santa Fe Institute: the authors of the two derivations it connects. Silence is not agreement, and none is claimed.

February to August 2026. The instrument: a blinded evaluation harness that catches its own failures. It found that unblinded scoring can flip a result's sign, and later that its own panel had scored measurement failure as success. Both findings were published. The attack that caught the second one now runs as a permanent regression test.

2026. The retraction, published against the author's own headline. The early sequential-scaling exponent of 2.24 was retracted after replication failed. The corrected figure, 0.49 with a bootstrap interval reaching below zero, replaced it in public. Every document carrying the old number was corrected. This row is the reason to trust the others.

2025 to 2026. Two working products, built alongside the research: Eden Legal AI, a bounded prototype of a litigation operating system, and Eden Control, in private beta.

Sources: track-record/index.html

The falsification posture

Every study in this programme states, in a timestamped public registration, the result that would refute it, before any data is collected. Most science tells you what it found. The registered programme page lets you watch science that cannot cheat: every prediction dated before its data, every way of being wrong written down first, nothing moved afterwards. The reader is not asked to trust the author. The reader is invited to watch him win or lose.

The honesty line at the head of the diligence table, quoted verbatim as it appears on the page:

Read this line before the table. Every row below is a PRIVATE DRAFT registration prepared and populated on OSF, awaiting the operator's human review and submission. NONE has been submitted or accepted yet: the programme's own rule is that the AI prepares drafts and only a human files. Any public sentence implying N registrations are FILED is false until submissions begin, and this dataset says so first.

The count as of 9 August 2026: 54 private drafts prepared, 4 final-ready, 50 working, 2 held under patent assessment and never uploaded, 0 submitted, 0 resulted.

The programme at a glance: 57 studies specified, written with gates in the private estate; 4 verified ready to register, packets machine-verified for manifests, hashes and randomisation; 0 registered on OSF; 2 held for disclosure review, withheld deliberately, pending a legal ruling.

One rule governs the page: a design appears there only after its registration is timestamped on OSF. Until then a study shows a name and a status and nothing else, because publishing a design ahead of its timestamp would spend the very thing a registration exists to protect. As registrations land, their rows fill in with the hypothesis, the kill condition and the OSF link, and nothing already shown is ever edited afterwards.

The kill routes are recorded in advance: the grid escape, the cross-class escape, the cross-class correction-exponent ratio, the moved half. One already fired. The early sequential-scaling exponent of 2.24 was retracted when replication failed, and the correction went on the public record. It was published against the author's own interest. That is the credential that the kill conditions are real.

Sources: research/registered-programme/index.html

What is asked and what it buys

The for-funders page states what a funder would and would not be funding at the level currently public.

You would be funding the instrument, not the theory. The measurement arm is named ARC-Beta-k. This programme has a quantitative stability criterion for self-improving systems on the table, prepared for registered measurement, with the honest state of every registration on the public diligence table. What the registered measurements, when they run, would decide: whether the stability ceiling, the reciprocal of the correction shortfall, survives contact with a real system at the value the independence assumption returns, whether β > k holds as a stability condition on a running corrector, and whether the conversion law U = I × R^α holds outside its stated model. If the registered measurements confirm it, it is the first stability criterion for self-improving systems computed from the corrector’s own measured scaling exponent, and the conversion law’s first measured exponent.

You would not be funding a confirmed result. The page is explicit: the mathematics hold inside a stated model and have not been tested on a real one. The empirical work is unregistered and unreplicated.

The reason to fund it anyway, in the page's own words: for someone with no institution behind him, the only question worth asking is whether he will tell you when it fails. The retraction was published against the author's own headline number: 2.24 down to 0.49, in public, with the working shown. Null results are published beside the positive ones. A defect in the evaluator that had been scoring measurement failure as a perfect result was found, published, and the attack that caught it now runs as a permanent regression test.

The programme's speculative layer, the long-range cosmological material, is quarantined from the ARC-Beta-k measurement programme. It contributes nothing to the claims, the registrations or the statistics. No measurement-programme work depends on it.

The costed plan is available on request. The public for-funders page carries the case for the instrument and the standing objections answered; the itemised budget is held for direct correspondence. An unaffiliated researcher has no supervisor, no ethics committee, and no co-authors to catch him. The only evidence available about his honesty is what he does when nobody is making him. That record is on the track-record ledger, dated, and it is the one thing on the funders page the author asks a reader to weigh.

Sources: research/for-funders.html

How to verify before trusting

The verify page lists five checks anyone can run against the programme. Two of them run in the reader's browser, against systems the author does not control. None of them needs the author's permission, and none of them needs the reader's trust.

Check 1. Hash what this site serves. Every load-bearing file has a published SHA-256. A tool on the page fetches the file the reader would be served and recomputes the hash in the browser, with WebCrypto, sending nothing anywhere. The December 2024 manuscript anchor has sent-side prefix f0d1f38f. The raw source objects are held privately; the hashes are public; source is available on request.

Check 2. Resolve the signing key. The December and April records verify under a DKIM key published in DNS at google._domainkey.mastermindpromotion.com. The reader asks Google's or Cloudflare's resolver for the key and compares fingerprints against the 2 August 2026 capture. A match corroborates the key currently served, from that vantage, today. It does not establish what the DNS record held in 2024, and it does not verify a signature by itself. That boundary is stated on the page before the button.

Check 3. Recompute a result. Clone the validation repository at github.com/MichaelDariusEastwood/arc-principle-validation and follow the run-it-yourself steps on the evidence spine. If a step fails, the claim fails. The programme DOI is 10.17605/OSF.IO/6C5XB. The paperback is ISBN 978-1-80605-620-0, published 2 January 2026, checkable in any library catalogue.

Check 4. Read what has been wrong. The corrections log is append-only and dated. It includes the retraction of the author's own headline number. A record with no corrections is a record nobody has checked.

Check 5. Try to kill a claim. Every claim carries a stated kill condition on the falsification dashboard, and every convergence links its primary source. One kill condition has already fired.

What a pass on any of these checks does not establish, in the page's own words: authorship, originality, and the merit of any argument are untouched by every check on this page. Hashes establish that bytes are unchanged; keys establish where signatures can be checked; neither establishes that the ideas are any good.

The standing offer: if a reader breaks a claim, the retraction goes on the record, publicly, with the reader's name on it and framing intact. A register that survives adversarial checking is worth something. There is only one way to find out.

Sources: verify/index.html