What this site does not claim

“He claims to have discovered three new laws of nature.”

The theory registration says what its laws are, in its own words:

What are called laws here are named conjectures under registered test. … Nothing here claims the standing of a law of nature, and this registration exists so that the claim can be earned or lost by measurement and by replication the author does not run.

The theory now names five laws, and none is presented as established: three were registered on 8 September 2026, and the two added on 24 September 2026 were registered as laws on 15 October 2026 in version 1.103 and have not been tested. Naming a proposed law is how it can be tested and killed, not a claim that it holds. What is claimed is that he stated, named and dated five conjectures, each graded on its own, and joined them in one theory.

Source: the theory registration, the statement paper.

“He claims the d/(d+1) exponent as his discovery.”

The paper that makes the argument says the opposite, in its section on what is new:

The claim is not that this formula is original, but that Cauchy's functional equations, combined with the space-filling and conservation conditions, explain why the earlier derivations, from different starting assumptions, arrive at the same form.

That section names the earlier work, beginning with West, Brown and Enquist in 1997, and the foundational paper says the same of itself: The ARC framework does not claim priority over the formula itself. What is claimed is the explanation. It has been compared against twenty-five empirical domains, with each domain's class fixed from known physics before fitting: the predicted family was preferred in nineteen on the original candidate set, and in eighteen on the rerun of 11 August 2026 that added the logarithmic candidate. The comparison is graded exploratory. The step that would make the exponent compulsory, a formal proof that its three conditions suffice, is stated as open; the origin-of-scaling-laws paper says of it: That proof does not yet exist. It is the most important open problem in this programme.

Source: the origin-of-scaling-laws paper (sections 11 and 12), the foundational paper, the dated prediction register.

“The theory is presented as proven, or as established science.”

The statement paper says it of itself, in its section on what it is not:

Not proved; nowhere here claimed to be.

The programme's distinctive feature runs the other way: it publishes what would kill each claim. The statement paper prints its kill conditions in section 7, exactly what would refute each claim, and the falsification report reproduces the papers' own refuting conditions, verbatim and hashed, in a dated snapshot. One published kill condition has already fired, against the programme's own early headline number, and the correction remains public. In the statement paper's words, the fastest honest way to kill this theory is to run it.

Source: the statement paper, the falsification report, the corrections page.

“He never admits error; he is the only validator of his own work.”

The first charge is false. The most striking number the programme ever produced stands publicly withdrawn, and the theory registration records it among its reversals, on purpose:

A theory-level registration that presented a clean sheet would be worth less than one that shows the pattern, and one that flattened three unlike reversals into a single word would be showing it badly. … A measured exponent of approximately 2.24 was published and then withdrawn on 20 March 2026 after it failed to replicate across architectures.

The other two reversals are a relation corrected because it ran backwards and an expression withdrawn and re-derived, and the corrections page keeps every material correction in the open.

The second charge is conceded as far as it is true: no independent attempt to replicate a result has yet been recorded, and the site says so. What the site does not do is count the author as a validator; his own reruns are not counted as confirmations. So the checking is handed to the reader. Two of the six checks on the verification page run in the reader's own browser, against systems the author does not control, and the test catalogue keeps a dated register of independent attempts, in which a failed replication counts as prominently as a success.

Source: the theory registration, the corrections page, the home page, verify, don't believe, run it yourself.

“He claims to have predicted Google's Willow result.”

The programme's register of its dated predictions grades Google Willow in one sentence, with both dates:

it was announced with its paper in Nature on 9 December 2024, and the team’s preprint was public from 24 August 2024, before the December manuscript record, so it is graded convergent timing and never prediction.

So the result was not predicted, and Paper II gives the reason: the December record cannot have predicted a result its own authors had announced beforehand. The author's manuscript of 8 December 2024 fixes the date of his own framework and nothing about Google's experiment; it falls a day before the announcement and months after the preprint. The register's census of every dated prediction carries the row as convergent timing only, never a prediction, and the claims ledger gives priority for the experiment, its apparatus and its figures to Google Quantum AI.

Source: the dated prediction register (section 3, and its census in section 8.4), Paper II (section 1.4), the claims ledger, record PC-006.

“He claims to be first to say that control of AI becomes impossible.”

The site concedes the point by name. Roman Yampolskiy argued in 2020 that advanced AI cannot be fully controlled (On Controllability of AI, arXiv:2008.04071), four years before the author's record of 8 December 2024, and the related-work page concedes it to him in full, in his own words. The theory's own premise is narrower, a limit on how long control can be assumed, and the statement paper words it so:

This is a limit claim, not an instant one … what no constraint can be is the thing that holds forever, so control cannot be assumed to persist, and the burden of proof sits with the claim that it will.

The priority claimed is for the whole, not its parts: the same paper concedes the raising vocabulary to Turing and De Kai, and says the claim is the first dated statement of the conjunction.

Source: Yampolskiy (2020), related work, the statement paper.

“There are no equations and nothing testable, only prose.”

The statement paper sets the laws out in symbols, and defines every term in prose:

In symbols, for the reader who wants them now, and freely skipped by everyone else (section 3 defines every term in prose): capability U = I × R^α, the exponent measured rather than assumed; stability while β > k, correction out-scaling drift; the ceiling α_crit = 1/(1 − γ); and the conjectured ARC Bound, α ≤ 2.

Each is graded by what has been shown: Law I's exponent estimate is inconclusive, Law II is proved in its minimal model only, and Law III is derived under stated assumptions, with its value of two conjectured. Law II's paper states its results as theorems, and fourteen of them can be recomputed in a few minutes from the verification page, where a pass checks the mathematics, not the world. Laws IV and V, added on 24 September 2026 and registered as laws on 15 October 2026 in version 1.103 and not yet tested, have their mathematics set out at version 1.1 of 7 October 2026. What is testable is on the record: the theory's twenty-two propositions were registered on 8 September 2026, one trial unit was registered on OSF on 1 October 2026 and has been public since 4 October 2026, and the others are drafts awaiting the author's submission. The test catalogue lists sixteen runnable tests, says what each one decides, and states that none of them is a deciding study.

Source: the statement paper, verify, don't believe, the registered programme, run it yourself.

“It is numerology: mystical significance read into constants.”

No measured claim of the programme draws its support from a numerical coincidence. The fine-structure constant does appear in the corpus, and saying otherwise would be false, but in three ways, none of them numerology. The first is a clash of letters, which the 2026 papers flag in a terminology note:

Alpha appears in this programme in two distinct meanings across eras: the fine-structure constant in the December 2024 formalism, and the ARC Bound exponent in the 2026 work.

The second is the speculative cosmology, HRIH. Its paper, and the page that introduces it, discuss the standard fine-tuning observations of physics; the paper gives the constant at its standard value, the bibliography lists NIST's published value, and the paper's own limitations section states that it does not predict cosmological parameters from first principles and does not derive the fine-structure constant. The third is the book's passages on fine-tuning, which sit in the same labelled speculation. That speculation is filed apart from the theory's laws, as the statement paper says in as many words: Not a cosmological proof: HRIH stands at its own rung. Nothing on any page of this site derives anything from the Planck scale; apart from this register's own denial, the phrase appears on none of them. The programme's measured exponent is printed with its uncertainty, and the most striking number it ever produced was withdrawn in public when replication failed to support it.

Source: Paper I's terminology note, the HRIH paper's limitations section, the bibliography, the statement paper.

“He claims institutional or academic authority he does not have.”

The statement paper says it about itself, unprompted:

The author of this paper holds no institutional affiliation, no doctorate, and no journal’s imprimatur, and the paper is built so that none is needed to check it.

It is as plain about review: no credentialed external human has yet reviewed this work, and this paper says so rather than implying otherwise. No paper in the programme is peer-reviewed. In place of borrowed authority the programme offers checkable structure: dated records, public registers, hashes a reader can recompute, and registered predictions. The reader is asked to verify, never to believe.

Source: the statement paper, the research hub, verify, don't believe.