Verify AI/ML benchmark claims against papers, official repos, model cards, evaluation scripts, and benchmark documentation. Use when benchmark tables, leaderboard positions, or "SOTA" claims affect the recommendation and need apples-to-apples verification.
Benchmark claims are only useful when their context is verified.
Decide whether a benchmark claim is directly comparable, partially comparable, or not comparable.
verification-protocol for a fuller verification trail.Use this order:
references/verification-checklist.md.assets/verification-notes-template.md for persistent notes.