Assess this programme

One question decides the century: when the thing we build exceeds us, what determines what it does? The ARC Theory made that question measurable. The eclipse measurement is the deciding experiment: an architecture-blind null whose expected ratio, under the specified exchangeability model, is exactly 1.00 with no free parameter to fit, a prediction materially above it, and the kill condition written in advance. Einstein was a frame until Eddington sailed; the expedition here is specified, written up as draft registrations awaiting human submission, and waiting. It does not need the author’s hours. It needs a funder’s decision.

No lab funding · no access agreements · no funder has ever shaped the questions. The full conflicts declaration stands further down as the receipt.

What stands here is not a confirmed theory. It is the instrument built to test one, and the record of how it behaves. Referee, reviewer, journalist or programme officer: this page is the audit room.

Before the instrument, the conduct record, because it is what any assessor is actually weighing. The headline exponent was left free on the December 2024 record and measured later; when replication returned roughly 0.49 (interval [-1.3, 2.9]) against his own printed book, the retraction was published and stands. Serious assessment does not reward certainty. It rewards a process that corrects, and this one has corrected in public, at cost, against its own strongest number.

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) (inherited from West, Brown and Enquist) 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, so it is a correctability measure and the correction target carries its own separate burden of proof.

Why this deserves scrutiny before anything else does: the programme has located no quantitative stability criterion for self-improving systems in the prior work, the thing that decides whether every other safeguard means anything, and this programme has one on the table, prepared for measurement under ARC-Beta-k, its measurement arm, with the honest state of every registration on the public diligence table: the theory registration of 8 September 2026, the first study unit registered on 1 October 2026, and the rest written as drafts awaiting submission.

What is claimed, at full strength: a conversion law for creation, U = I × R^α, capability as intelligence converted through recursion; a stability ceiling that is the reciprocal of the cap on how well any internal corrector can scale, which under the same-class independence assumption returns two; and one accumulation law underneath both, the exponent of the conversion law and the ceiling of the correction law, replacing an earlier collapsed-identity framing withdrawn as vacuous in the dated record. If the measurements, run under those registrations once submitted, confirm it, it is the first stability criterion of this class the programme has located, computed from the corrector’s own measured scaling exponent, and the conversion law’s first measured exponent on the programme’s own record. The ways it can be wrong are its armour, written in advance: the grid escape, the cross-class escape, the cross-class correction-exponent ratio, the moved half. And the author’s public retractions are the credential that the kill-conditions are real: three sit on the corrections log, the α ≈ 2.24 headline retraction was the kill-condition that fired, and every one was published against his own interest.

The record runs as a dated ladder: the manuscript record of 8 December 2024 (published with that of 30 April 2025 on OSF, DOI 10.17605/OSF.IO/QASRK) gives the equation and the definitions of all three terms; the dated record of 30 April 2025 carries the squared form for the first time, as the framework equation, with the stability-ceiling derivation added later in 2026; the printed book of 2 January 2026 gives the exponent as a numbered testable prediction, the author’s printed prior statement, never called a preregistration because it carries no protocol, and read naively contradicted by the measured 0.49 (interval [-1.3, 2.9]), surviving only as the stability-frontier law; the 2026 programme, ARC-Beta-k, gives the mechanism and the meters; the draft registrations, once submitted and run, give the verdict. Where each named source stops, line by line →

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 six-element joint pattern and the scaling relation were not.

First, the output ledger: the track record →

The strongest thing here is methodological. Runnable, falsifiable measurements for questions that normally get argued rather than tested. An evaluator built to fail loudly instead of quietly. Draft registrations written, dated and prepared before the data exists, awaiting human submission.

What is not claimed

A confirmed result. The mathematics hold inside a stated model and have not been tested on a real one. No study unit registered on OSF has yet reported a result, no run has yet been repeated outside the programme, and each earlier run is dated from its own evidence. The related work page sets out in full where the components have prior work and where the joint arrangement exceeds it.

The question that decides it

Here is the part that matters. For someone with no institution behind him, it is the only question worth asking. Will he tell you when it fails?

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.

If your job is to evaluate rather than explore, the independent scientific review page is the control panel: claim status, kill routes, the deciding experiment and the two-second verification, in the calibrated register.

One claim you can check tonight, without trusting me:

git clone https://github.com/MichaelDariusEastwood/arc-principle-validation
cd arc-principle-validation/papers/Paper-X-Coupled-CoScaling-Correction
python3 -m venv .venv && . .venv/bin/activate
pip install -r requirements.txt && python code/test_theorems_independent.py

I publish the null results beside the positive ones. I found a defect in my own evaluator. It had been scoring measurement failure as a perfect result. I published that too, along with the attack that caught it.

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 above. It is dated. It is the only thing on this page I would ask you to weigh.

Conflicts of interest, declared

This programme takes no funding, employment or access agreement from any AI company; models are used at retail rates. No funder has ever shaped this programme’s questions.

→ the corrections log   → the prior work boundary   → what would falsify each claim

Prefer one document? The assessment brief, as one print report.