Run 820f682 · Own gold set (private corpus)

Run 820f682 Own gold set (private corpus), recorded 26 June 2026, 22:00:37 UTC.

dirty tree · not cited
determinism ok
private corpus · not reproducible

Benchmark metrics
Metric	Value	Definition
Recall@5	0.9554	Mean over questions of \|evidence ∩ top-k\| / \|evidence\|.
nDCG@5	0.9285	With binary relevance.
MRR	0.9333	Mean of 1/rank of the first evidence session.
Precision@5	0.2143	Mean over questions of \|evidence ∩ top-k\| / k. It can't reach 1 when a question has fewer than k evidence sessions, so it is shown on receipts only.

Conditions
Queries	56
k	5
Strategy	bm25
Gold path	benchmarks/recall-gold.json
Tests ran	715
Tests failed	0
Environment	Hardware, OS and Python version: not recorded for this run.
Ledger row	133

Raw row

{"ablation": null, "checks": {"determinism_ok": true, "index_validate_ok": true, "test_pass_rate": 1.0, "tests_failed": 0, "tests_ran": 715}, "commit": "820f682", "committed": "2026-06-26T21:32:55Z", "corpus_fingerprint": "e0468927da74896f05f0feb956d4a71bfd633a76993d98ff78d0fd0a03dcacbc", "corpus_size": 193, "dirty": true, "gold_path": "benchmarks/recall-gold.json", "metrics": {"context_tokens_mean": 3998.8928571428573, "latency_ms_mean": 10.342273892742273, "latency_ms_p50": 10.171520501899067, "mrr": 0.9333333333333333, "ndcg_at_k": 0.9285053385997024, "precision_at_k": 0.21428571428571427, "recall_at_k": 0.9553571428571429}, "mode": "eval", "num_queries": 56, "reliability": 98.21, "run": "2026-06-26T22:00:37Z", "schema": "agentware-bench/v1", "strategy": "bm25", "subject": "Merge feat/260625-kb-mcp-server", "suite": null, "top_k": 5}

reliability is an internal composite health score (0–100), not a benchmark metric.