One subdir benchmark for the find_decisions MCP tool via rac find <query> corpus/ --decisions --json. This tool — the one with a structural liveness defense — previously had zero benchmark coverage anywhere; supersession was tested only through search_artifacts, which applies no liveness filter.
Fixture corpus (FDB- ids): supersession chains where the superseded member is the lexically best match for the query, plus deprecated, proposed, and non-decision distractors. Cases assert the live decision is returned AND retired ids never appear anywhere in the full returned list. Metrics: P@k, R@k, MRR, negative_violations over the full list; gate negative_violations == 0 plus calibrated floors.
Roadmap: rac/roadmaps/benchmarks/tool-benchmarks.md in rac-core [roadmap:tool-benchmarks].
One subdir benchmark for the
find_decisionsMCP tool viarac find <query> corpus/ --decisions --json. This tool — the one with a structural liveness defense — previously had zero benchmark coverage anywhere; supersession was tested only throughsearch_artifacts, which applies no liveness filter.Fixture corpus (
FDB-ids): supersession chains where the superseded member is the lexically best match for the query, plus deprecated, proposed, and non-decision distractors. Cases assert the live decision is returned AND retired ids never appear anywhere in the full returned list. Metrics: P@k, R@k, MRR,negative_violationsover the full list; gatenegative_violations == 0plus calibrated floors.Roadmap:
rac/roadmaps/benchmarks/tool-benchmarks.mdin rac-core [roadmap:tool-benchmarks].