All 6 of 6 query classes A-G from §1.4 (excluding Phase 1.7 which is its own design) now have documented baselines on the v1.2.0 candidate DB:
| Baseline | Class | Headline | Interpretation |
|---|---|---|---|
search-quality-baseline-v1.2.0.md | A + B | MRR 0.9467, P@1 46/50 | Strong; the canonical use case |
search-quality-deprecation-baseline-v1.2.0.md | E | 30/30 Swift wins, p = 0.0078 | Strong; anti-hallucination axis intact |
search-quality-crosssource-baseline-v1.2.0.md | F | 19/19 conditional, p = 1.9 × 10⁻⁶ | Strong but biased (intentionally) |
search-quality-fragment-baseline-v1.2.0.md | D | P@1 = 1.0, P@5 = 0.92 | Strong; symbol_components working |
search-quality-acronym-baseline-v1.2.0.md | C | 4/22 = 18% | Weak; synonyms not at ranking time |
search-quality-prose-baseline-v1.2.0.md | G | 4/15 = 26.7% strict (53-67% adj) | Methodology-limited; BM25F trade-off |
search-quality-symbol-attribute-baseline-v1.2.0.md (this doc) | H | P@5 = 0.25 | Weak; default path doesn't consult doc_symbols metadata |
Remaining: §14.4 Phase 1.7 anti-hallucination agent-end-to-end eval. That is its own design (docs/design/anti-hallucination-eval.md, not yet written) and the actual release-blocker test.