Datasets:
Contributing review evidence
Start with docs/EXPERT_REVIEW.md or docs/AI_AGENT_GUIDE.md. Use claim IDs from audit/claim_ledger.json and the structured report in review/.
For a correctness issue, supply the exact theorem, missing inference or reproducible counterexample, and minimal correction. For an implementation issue, supply the failing command, environment, input hashes, and a patch. For an originality issue, supply a primary source and the mathematical translation to the present claim.
Keep new runs under runs/; preserve the original reports and archived research package. Changes to the mathematics require a research-version update and affected-claim list. Changes only to documentation or packaging require a packaging-version update. Regenerate the manifest for a changed release and record the modifications.
Submit a discussion or change proposal through the repository's available community controls after publication. Do not claim a proposal is accepted before the maintainer records acceptance. Human and AI reviews must remain distinguishable, and no review creates an endorsement beyond its stated scope.
Check LICENSE_STATUS.md before redistribution or using the work as a dependency. A future declared license must be an explicit maintainer decision; contributors should not silently relicense the preserved package.