What happened
OpenAI says the arguments were generated by Astra while the system was still in development, then prepared as manuscripts with the same model and formalized in Lean (openai/ten-proofs; the project reports zero sorries). Pre-release and public Astra are the same named release. The ten items are: high-dimensional sphere-packing bounds down to the Cohn–Elkies threshold; exponentially improved binary and spherical code bounds; an explicit finitely presented non-sofic group; a counterexample to Connes’s rigidity conjecture; permanent arithmetic-circuit and formula lower bounds including Ω(n^4/log n) for formulas; exponential parallel repetition for general two-player quantum games; polynomial-factor CVP hardness; the sharp Ehrhart volume inequality in every dimension; a superexponential lower bound resolving Erdős 183 on multicolor triangle Ramsey numbers; and resolutions of the compactness and degeneracy conjectures (Erdős 146 and 180). This is scored as one Historic cluster because the theorems would have been Major or Historic as independent discoveries; bundling them in one paper is not a 10-point ceiling. It is not the September Navier–Stokes search system, which OpenAI described as more capable than Astra. Independent reconstruction of the full set remains incomplete, so the event stays provisional.