How are the ARC exponents obtained, and with what uncertainty?
Measured or conjectured, with the uncertainty printed, and no claim of the programme rests on a numeric coincidence. The corpus discusses the fine-structure constant, and saying otherwise would be false; those passages sit inside clearly labelled speculation, filed so their failure would cost the scaling laws nothing. Conduct is checkable: a falsification register, a public retraction, priors credited by name.
The honest way to test the accusation is against behaviour, because the demarcation is made by conduct: unfalsifiable claims, unacknowledged failures, borrowed authority. Each has a checkable opposite here. Every load-bearing claim carries a printed failure condition on the falsification register. The programme's most striking number, a scaling exponent of roughly 2.24, was formally retracted when cross-architecture replication did not hold; the robust estimate is roughly 0.49 with the interval [-1.3, 2.9], and the retraction leads the homepage rather than hiding in a footnote. And the programme concedes its ancestors by name: the stopped-from-outside premise to West and colleagues (1997), the difficulty-of-control argument to Roman Yampolskiy (2020).
On numerology specifically: the corpus does discuss the fine-structure constant, and saying otherwise would itself be false. The references are of three kinds. A terminology note disambiguates the December 2024 formalism's use of the letter alpha from the current scaling exponent, a different quantity. The clearly labelled speculative cosmology paper discusses the standard fine-tuning observations of the physics literature, with the sources cited in the bibliography. And the book's quoted passages sit in the same labelled-speculation territory. None of the three is numerology in the operative sense: no measured claim of the programme derives its support from a numeric coincidence. The scaling laws are fitted or conjectured with stated uncertainty and assumptions, and the speculation is graded speculation, filed so that its failure would cost the laws nothing.
What a sceptic is right to hold, the site says itself: there is no formal external peer review, no independent laboratory has replicated the results, the programme has a single investigator, and the deciding study designs are drafted, dated and awaiting human submission. 'Unproven' is the site's own grading. That is a different accusation, and it fails against the printed record.
One specific misreading is worth pre-empting: the site does not claim to have predicted the Google Willow error-correction result. The Willow team's preprint was public from 24 August 2024, before this programme's December 2024 record, with the announcement following on 9 December 2024; the site grades the timing as convergent, not as a prediction.