On November 17, Anthropic described two black-hole physics results involving Claude Fable 5.2. One concerns the Hawking–Page transition at weak coupling; the other establishes the exact limit of a simplified modular-bootstrap approach. The company states that neither resolves quantum gravity. Kaldata discussed the work on November 20.
The linked Hexagon preprints were deposited on November 7. They expose the specific mathematical formulations and restrictions. The announcement date should not become the date of every underlying calculation, and a preprint’s presence does not establish completed independent peer review.
Assessing the scientific claim
Our editorial conclusion is to identify the problem actually solved first. A theoretical result can matter within a precisely defined model without describing every real black hole. Omitting that condition transforms a specific advance into a much broader, unsupported promise.
The second question is what a checking procedure actually establishes. Correct mathematical steps, applicable assumptions and physical interpretation are related but different tasks. Readers benefit from knowing which is supported and which remains a research question.
A proposal for responsible research agents
We would require a short map for each proposed idea: problem statement, assumptions, derived claim and checking method. Unverified portions should remain visible. This would also help when several agents work concurrently and incorporate one another’s results.
We also propose separately reproducing decisive steps through a person or an independent tool, without accepting the conclusion in advance. Similar answers from copies of one model may supply a useful lead but cannot alone rule out a shared mistake. SIExpo has not reproduced the physics calculations and does not certify the proofs.
A useful research record should include failed directions too. Otherwise it is impossible to assess the attempts, time and human effort behind the successful account. For a future scientific team, that is an important part of deciding where automation fits.
We would preserve the connection between a checked statement and the version of the document that contains it. A later revision can clarify a definition or alter a condition. Treating every version as interchangeable makes both criticism and reproduction harder, even if the central result survives unchanged.
The news here is the availability of specific reported results and accessible research material. The strongest reading preserves the mathematical conditions, actual dates and clear distinction between contributing to a method and delivering a final theory of nature. That distinction supports informed interest without requiring readers to accept an exaggerated account of scientific completion.



