OpenAI two-target audit - wave 1 evidence volume (September NS certificate repository adjudication; QRH proven theorem chain two-face closure)
Independent wave-1 audit of two OpenAI claims. Target 1: the September Navier-Stokes/Euler Lean certificate repository (openai/NavierStokesAndEuler) - the first independent verbatim adjudication of the critique in arXiv 2610.08144: both partly right (the flagged proof-chain estimates differ from the NL paper and are load-bearing; the proved theorem matches the NL Theorem 1.1 clause-by-clause; the forced blow-up falls literally within Clay statements (C)/(D)). Target 2: the QRH proven theorem chain in openai/math (Nonvanishing.lean) - the challenge-to-proven correspondence closes verbatim with zero shadowing and a 2,924-module sorry-free, axiom-clean import closure; the rebuild-is-verification narrative holds on this challenge face. Scope honestly bounded: no compilation, no kernel replay, no certificate numerics. Companion to repository issue #23 of hy7pc8gfmf-dotcom/rocqwash (AGPL/ICLA protocol; AI-assisted by design, human-gated filing).
Authors
- 宝军 王 (ORCID: https://orcid.org/0009-0001-4240-9045)
- 雪峰 高 (ORCID: https://orcid.org/0009-0007-9904-1599)
- Zhou (ORCID: https://orcid.org/0009-0001-3206-5824)
- 光照 祖 (ORCID: https://orcid.org/0009-0006-7811-6638)
- 夏 挽岚
- Zeng (ORCID: https://orcid.org/0009-0005-9811-5598)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24410025
- Primary Topic
- Logic, programming, and type systems
- Type
- article
- Field-Weighted Citation Impact
- 0.00