# Authorship, AI use and review ## Human author **Will Strinz** is the sole human author and the person responsible for this repository. He directed the work, made the decisions recorded here (scope, licence, what to release) and is accountable for the release. He is not an expert in Euclidean Ramsey theory, and he has not reviewed the proof of Theorem D line by line. ## AI contributions, by name | Contributor | What it did | |---|---| | Claude Opus 5.5 (Anthropic), in claude.ai | The exploratory conversation from which the project started | | **Claude Opus 5.5** (Anthropic), in Claude Code | Coordination of the work; the Lean proof (`RadianRamsey/`, first written at Lean v4.34.1 as rope AV3 of spherical-ramsey); the exact determinant check used while exploring (spherical-ramsey's `certificates/radian_det.py`); the port to the Lean module system and this package (challenge, controls, CI, documentation); the reviews listed below as AI reviews | | **Claude Sonnet 5.5** (Anthropic), in Claude Code | The prior-work and novelty searches (`docs/PRIOR-WORK-SEARCH.md`); a fresh reproduction of rope AV3 of spherical-ramsey (this proof at Lean v4.34.1) from its README alone | The agents worked in Claude Code. Every machine check reported in this repository was run by these agents or by GitHub-hosted CI that they set up; the author did not re-run them. ## Foundation **OpenAI's result 172**, *A classification of finite Euclidean Ramsey configurations*, and its complete Lean proof (github.com/openai/math, commit `fd4aeeb2`), is the foundation of this work. Theorem D applies the argument of their Proposition 7.6 to a one-parameter family, and the Lean proof here builds on OpenAI's Lean proof, whose 35 files used here are included (Apache-2.0; see `NOTICE`). ## Review: who checked what | Check | By | Kind | |---|---|---| | Lean proofs: kernel checking, axioms, statement match, refusal of five negative controls (this package, Lean v4.35.0-rc2) | Lean, NanoDa and con-ron kernels through the toolchain's `lake comparator`, on a fresh GitHub runner | machine | | The same proof at Lean v4.34.1 (rope AV3 of spherical-ramsey): statement match against a statement-only challenge, OpenAI's own challenges, and five negative controls | stock Comparator (Lean kernel, Landlock sandbox), on the builder's machine and on a fresh GitHub runner | machine | | Fidelity of rope AV3's statements (its `FIDELITY.md`, the basis of `docs/FIDELITY.md` here), by an agent that did not build it | Claude Opus 5.5, read-only | AI review: faithful, fixes to the documentation only, all applied | | Fresh reproduction of rope AV3 from its README alone, by an agent that did not build it | Claude Sonnet 5.5 | AI reproduction: reproduced (exit 0, all ten checks as expected); its friction notes were fixed in that README | | The port (`docs/PORT-LOG.md`): header-only changes to OpenAI's files; header, import, namespace and comment changes only to this project's files | mechanical diffs (`REPRODUCE.md` §5 for OpenAI's files) and Comparator's statement and definition match | machine | | This package's statements and documentation (`Challenge.lean` against the rope's challenges, the proof code against the rope's, the OpenAI files against mobius-ramsey's, `formalization.yaml`, the controls, the README and the metadata), by an agent that did not build them | Claude Opus 5.5, read-only | AI review: statements identical to the rope's, faithful; documentation fixes only, all applied | | The proof sketch of Theorem D and the θ = 1 corollary (spherical-ramsey's note) | none | not separately reviewed: the AI fidelity review of rope AV3 covered the Lean statements, not the note's sketch; the Lean proof is the machine-checked argument | | Novelty of Theorem D | Claude Sonnet 5.5 (arXiv, OpenAlex, zbMATH, GitHub, MathOverflow searches) | AI search; not a literature review by an expert | **No human expert has reviewed this work.** The θ = 1 corollary rests on the Lindemann–Weierstrass theorem, which is cited, not formalized.