{
  "benchmark_track": "external_public_rag",
  "dataset": "nvidia/TechQA-RAG-Eval",
  "notes": [
    "This comparison is local and deterministic. It does not claim a provider-backed embedding result.",
    "The primary public retriever remains a local TF-IDF system so the public benchmark is reproducible in CI."
  ],
  "primary_retriever": {
    "case_count": 480,
    "description": "Local TF-IDF document retrieval with title boosts and exact phrase scoring.",
    "failed_case_count": 197,
    "label": "Local TF-IDF public retriever",
    "metrics": {
      "abstention_accuracy": 0.7875,
      "answerable_false_abstention_rate": 0.0443,
      "impossible_abstention_rate": 0.1146,
      "mean_reciprocal_rank_at_3": 0.7617,
      "retrieval_hit_rate_at_3": 0.8073,
      "top1_citation_accuracy": 0.724
    },
    "system_id": "local_tfidf_public_retriever"
  },
  "report_type": "techqa_public_retriever_comparison",
  "status": "evaluated",
  "summary": {
    "baseline_system_id": "keyword_title_baseline",
    "impossible_abstention_rate_delta": -0.2812,
    "primary_system_id": "local_tfidf_public_retriever",
    "retrieval_hit_rate_at_3_lift": 0.1563,
    "system_count": 2,
    "top1_citation_accuracy_lift": 0.1719
  },
  "systems": [
    {
      "case_count": 480,
      "description": "Simple title-token overlap baseline over public support documents.",
      "failed_case_count": 253,
      "label": "Keyword title baseline",
      "metrics": {
        "abstention_accuracy": 0.6625,
        "answerable_false_abstention_rate": 0.2708,
        "impossible_abstention_rate": 0.3958,
        "mean_reciprocal_rank_at_3": 0.5968,
        "retrieval_hit_rate_at_3": 0.651,
        "top1_citation_accuracy": 0.5521
      },
      "system_id": "keyword_title_baseline"
    },
    {
      "case_count": 480,
      "description": "Local TF-IDF document retrieval with title boosts and exact phrase scoring.",
      "failed_case_count": 197,
      "label": "Local TF-IDF public retriever",
      "metrics": {
        "abstention_accuracy": 0.7875,
        "answerable_false_abstention_rate": 0.0443,
        "impossible_abstention_rate": 0.1146,
        "mean_reciprocal_rank_at_3": 0.7617,
        "retrieval_hit_rate_at_3": 0.8073,
        "top1_citation_accuracy": 0.724
      },
      "system_id": "local_tfidf_public_retriever"
    }
  ]
}
