{
  "schema_version": "1.0",
  "id": "retrieval-ranker-evaluation-2026-10-28",
  "title": "BM25-style ranking versus TF-IDF on local technical documentation",
  "published_at": "2026-10-28",
  "evidence_status": "preliminary reproducible evaluation",
  "corpus": {
    "documents": 52,
    "queries": 21,
    "source": "Installed public system-documentation files; no private user or project corpus."
  },
  "aggregate_metrics": {
    "BM25": {
      "precision_at_5": 0.390476,
      "recall_at_5": 0.746032,
      "MRR": 0.789683,
      "nDCG_at_5": 0.674508,
      "median_query_ms": 0.2031
    },
    "TF_IDF": {
      "precision_at_5": 0.27619,
      "recall_at_5": 0.531746,
      "MRR": 0.638889,
      "nDCG_at_5": 0.433675,
      "median_query_ms": 0.1448
    }
  },
  "verification": {
    "exact_top_five_ordering_matches": "42/42",
    "independently_recomputed_metric_values": "168/168",
    "source_span_integrity_failures": 0,
    "held_out_paraphrase_gate": "PASS within preregistered tolerance",
    "run_checksums": "verified"
  },
  "limitations": [
    "Relevance grades are researcher-curated source references, not independent human judgments.",
    "Independent code-path agreement does not validate the relevance labels, representativeness or claim entailment.",
    "Not production validated; do not infer that BM25 is universally superior."
  ],
  "artifacts": {
    "report": "https://darkoza.com/science/reports/retrieval-rankers/report.md",
    "csv": "https://darkoza.com/science/reports/retrieval-rankers/results.csv",
    "raw_results": "https://darkoza.com/science/reports/retrieval-rankers/results.json",
    "automated_gate": "https://darkoza.com/science/reports/retrieval-rankers/automated-gate.md",
    "held_out_audit": "https://darkoza.com/science/reports/retrieval-rankers/heldout-audit.md"
  }
}