Single-token canonical-lookup queries (Class A) + framework-root queries (Class B) from scripts/eval/search-quality-phase1.py's CANONICAL_QUERIES corpus. For each query: cupertino search --search-db <db> --format json --limit 10 "<query>", parse the URI list, check each against the per-query right-answer regex, record the first matching rank. Per-query reciprocal rank = 1 / first_rank (0 if no match in top 10). NDCG@10 uses gain=1 per match, IDCG=1 (multi-match sibling URIs can push NDCG > 1 as documented in docs/design/search-quality-eval.md §8.2; this is preserved here for comparability with the v1.0.2 audit).