# AI-use statement

**Paper:** Strongly regular graphs on 160, 232, 261 and 280 vertices from fusions of orbitals of PΓL(2,8) and 2^4:(S3×S3)

**Author and sole contributor:** Anton Kovaľ (ORCID 0009-0009-3182-6626), VÚB banka, Bratislava, Slovakia. No AI system is a contributor or an author.

## How AI systems were used

Taken from the paper's acknowledgements and verification sections, which state these facts in full.

- Claude (Anthropic) built the search image with its controls (one Claude Code session); a coordinating session re-verified the server hits with independent checkers; an analysis session rebuilt the group actions from scratch, enumerated all fusions of orbitals of the same actions (which gave three graphs the search had not reported), found the explicit cross-ratio rules, the group Γ₈ and the grid-and-flags model, computed the invariants and ran the family searches for q = 4, 16, 32; a further session wrote the paper and, as a check of the text, re-implemented the constructions from its wording (Appendix B). These sessions used the same model family, so their agreement is evidence of correct computation but not an independent mathematical review.
- The first version was audited blind by a session of a different model (Claude Fable), which re-implemented every construction and computation from the prose of the analysis note with its own code, confirmed all eight graphs and all structural and family claims, decided the Hoffman cocliques of the 261 and 280 graphs, and listed the corrections made in the second version.
- The second version was refereed by a further Claude Fable session with its own code written from the text of the paper; it re-derived the hand proofs, reproduced the computations, confirmed the coclique counts with a third program, and asked for the corrections made in version 3, which is the version deposited.

## Role of the author

The author directed the work, coordinated the AI sessions, checked the results and the mathematics, and is responsible for all content. No step of the argument is accepted on the judgement of an AI system: each claim is either proved in the text or is a finite computation that can be rerun from the deposited data archive (Zenodo, DOI 10.5281/zenodo.23313399).

## Records in this archive

These are the written outputs (reports) of the analysing, auditing and refereeing AI sessions, copied unchanged from the author's working repository. File paths cited inside them are repository-relative; most of the cited scripts and logs are in the Zenodo data archive of the paper. Full session transcripts are not included.

- `new_srgs_twoorbit_2026_10_11.md`: analysis note (Claude), the basis of version 1
- `audit_twoorbit8_fable_2026_10_12.md`: blind audit of version 1 (Claude Fable)
- `referee_new_srgs_twoorbit_v2_fable_2026_10_12.md`: referee report on version 2 (Claude Fable)
