Run 5b9e4b4 · Own gold set (private corpus)

Run 5b9e4b4 Own gold set (private corpus), recorded 24 June 2026, 23:00:08 UTC.

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

Benchmark metrics
Metric	Value	Definition
Recall@5	0.9911	Mean over questions of \|evidence ∩ top-k\| / \|evidence\|.
nDCG@5	0.9506	With binary relevance.
MRR	0.9470	Mean of 1/rank of the first evidence session.
Precision@5	0.2214	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	113
Tests failed	0
Environment	Hardware, OS and Python version: not recorded for this run.
Ledger row	20

Raw row

{"ablation": null, "checks": {"determinism_ok": true, "index_validate_ok": true, "test_pass_rate": 1.0, "tests_failed": 0, "tests_ran": 113}, "commit": "5b9e4b4", "committed": "2026-06-24T19:25:06Z", "dirty": true, "gold_path": "benchmarks/recall-gold.json", "metrics": {"context_tokens_mean": 3746.964285714286, "latency_ms_mean": 2.81057657204136, "latency_ms_p50": 2.7581874965108, "mrr": 0.9470238095238095, "ndcg_at_k": 0.9505981548991542, "precision_at_k": 0.22142857142857145, "recall_at_k": 0.9910714285714286}, "mode": "eval", "num_queries": 56, "reliability": 99.64, "run": "2026-06-24T23:00:08Z", "schema": "agentware-bench/v1", "strategy": "bm25", "subject": "feat(agent): failure handling with web search and knowledge base lookups", "top_k": 5}

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