{
  "release": "graphrag-v7-frozen",
  "graph_dev": {
    "questions": 1046,
    "source_checked_count": 1046,
    "all_facts_supported_count": 1046,
    "strict_count": 1046,
    "strict_accuracy": 1.0,
    "error_count": 0,
    "value_score": 1.0,
    "entity_score": 1.0,
    "relation_score": 1.0,
    "property_score": 1.0,
    "relation_choice_score": 1.0,
    "document_recall": 1.0,
    "source_recall": 1.0,
    "binding_score": 1.0,
    "path_score": 1.0,
    "answerability_accuracy": 1.0,
    "latency_mean_seconds": 0.081848,
    "latency_median_seconds": 0.071172,
    "latency_p95_seconds": 0.113998
  },
  "graph_new_test": {
    "questions": 80,
    "source_checked_count": 80,
    "all_facts_supported_count": 80,
    "strict_count": 80,
    "strict_accuracy": 1.0,
    "error_count": 0,
    "value_score": 1.0,
    "entity_score": 1.0,
    "relation_score": 1.0,
    "property_score": 1.0,
    "relation_choice_score": 1.0,
    "document_recall": 1.0,
    "source_recall": 1.0,
    "binding_score": 1.0,
    "path_score": 1.0,
    "answerability_accuracy": 1.0,
    "latency_mean_seconds": 0.093012,
    "latency_median_seconds": 0.091794,
    "latency_p95_seconds": 0.127876
  },
  "graph_historical_regression": {
    "questions": 320,
    "source_checked_count": 320,
    "all_facts_supported_count": 320,
    "strict_count": 320,
    "strict_accuracy": 1.0,
    "error_count": 0,
    "value_score": 1.0,
    "entity_score": 1.0,
    "relation_score": 1.0,
    "property_score": 1.0,
    "relation_choice_score": 1.0,
    "document_recall": 1.0,
    "source_recall": 1.0,
    "binding_score": 1.0,
    "path_score": 1.0,
    "answerability_accuracy": 1.0,
    "latency_mean_seconds": 0.081877,
    "latency_median_seconds": 0.078039,
    "latency_p95_seconds": 0.119541
  },
  "model_dev": {
    "questions": 46,
    "correct": 46,
    "accuracy": 1.0,
    "errors": 0,
    "first_pass_correct": 45,
    "format_repair_attempted": 1,
    "corrective_request_count": 1,
    "context_repair_attempted": 0,
    "normalized_first_response": 0
  },
  "model_new_test": {
    "questions": 80,
    "correct": 80,
    "accuracy": 1.0,
    "errors": 0,
    "first_pass_correct": 77,
    "format_repair_attempted": 1,
    "corrective_request_count": 3,
    "context_repair_attempted": 2,
    "normalized_first_response": 0
  },
  "model_historical_regression": {
    "questions": 320,
    "correct": 319,
    "accuracy": 0.996875,
    "errors": 0,
    "first_pass_correct": 310,
    "format_repair_attempted": 3,
    "corrective_request_count": 10,
    "context_repair_attempted": 7,
    "normalized_first_response": 0
  },
  "model_new_test_by_category": {
    "candidate_set_aggregation": {
      "questions": 7,
      "correct": 7,
      "accuracy": 1.0
    },
    "conjunctive_disambiguation": {
      "questions": 6,
      "correct": 6,
      "accuracy": 1.0
    },
    "document_author_category_multi_hop": {
      "questions": 2,
      "correct": 2,
      "accuracy": 1.0
    },
    "negative_lookup": {
      "questions": 8,
      "correct": 8,
      "accuracy": 1.0
    },
    "person_keyword_title": {
      "questions": 10,
      "correct": 10,
      "accuracy": 1.0
    },
    "person_multiple_categories": {
      "questions": 7,
      "correct": 7,
      "accuracy": 1.0
    },
    "person_title_year": {
      "questions": 10,
      "correct": 10,
      "accuracy": 1.0
    },
    "same_academic_year_pair": {
      "questions": 10,
      "correct": 10,
      "accuracy": 1.0
    },
    "same_start_year_pair": {
      "questions": 10,
      "correct": 10,
      "accuracy": 1.0
    },
    "shared_keyword_pair": {
      "questions": 10,
      "correct": 10,
      "accuracy": 1.0
    }
  },
  "model_regression_by_category": {
    "candidate_set_aggregation": {
      "questions": 32,
      "correct": 32,
      "accuracy": 1.0
    },
    "conjunctive_disambiguation": {
      "questions": 40,
      "correct": 39,
      "accuracy": 0.975
    },
    "document_author_category_multi_hop": {
      "questions": 24,
      "correct": 24,
      "accuracy": 1.0
    },
    "negative_lookup": {
      "questions": 32,
      "correct": 32,
      "accuracy": 1.0
    },
    "person_keyword_title": {
      "questions": 32,
      "correct": 32,
      "accuracy": 1.0
    },
    "person_multiple_categories": {
      "questions": 24,
      "correct": 24,
      "accuracy": 1.0
    },
    "person_title_year": {
      "questions": 24,
      "correct": 24,
      "accuracy": 1.0
    },
    "same_academic_year_pair": {
      "questions": 40,
      "correct": 40,
      "accuracy": 1.0
    },
    "same_start_year_pair": {
      "questions": 32,
      "correct": 32,
      "accuracy": 1.0
    },
    "shared_keyword_pair": {
      "questions": 40,
      "correct": 40,
      "accuracy": 1.0
    }
  },
  "protocol": {
    "seed": 20261026,
    "dev_count": 1046,
    "fresh_test_count": 80,
    "test_categories": {
      "shared_keyword_pair": 10,
      "same_academic_year_pair": 10,
      "same_start_year_pair": 10,
      "person_title_year": 10,
      "person_keyword_title": 10,
      "document_author_category_multi_hop": 2,
      "person_multiple_categories": 7,
      "candidate_set_aggregation": 7,
      "conjunctive_disambiguation": 6,
      "negative_lookup": 8
    },
    "test_positives": 72,
    "test_negatives": 8,
    "available_groups": {
      "shared_keyword_pair": 1883,
      "same_academic_year_pair": 535,
      "same_start_year_pair": 153,
      "person_title_year": 30,
      "person_keyword_title": 146,
      "document_author_category_multi_hop": 2,
      "person_multiple_categories": 7,
      "candidate_set_aggregation": 54,
      "conjunctive_disambiguation": 628,
      "negative_lookup": 106
    },
    "old_gold_corrections": 10,
    "regression_count": 320,
    "llm_dev_count": 46,
    "target": 0.92,
    "policy": "Freeze after complete DEV; run fresh TEST once, retain every failure. Regression remains separate.",
    "annotation": "Independent raw-source oracle plus AI source review; not human review."
  },
  "tests": {
    "passed": 591,
    "skipped": 0
  },
  "source_review": {
    "sample": 20,
    "supported": 20,
    "reviewer": "AI direct raw Markdown review, not human"
  },
  "functional_scenarios": {
    "passed": 9,
    "total": 9
  },
  "historical_rag_replay": {
    "passed": 5,
    "total": 5,
    "new_generation": false
  },
  "failed_test_ids": [],
  "failed_regression_ids": [
    "wide-285"
  ],
  "limitations": [
    "Known corpus, persons and template families; new query groups, not unseen profiles.",
    "Source oracle plus AI review; no human annotation.",
    "Only two fresh document-author/category groups and seven fresh category-list groups remain.",
    "End-to-end native OpenWebUI tool routing is point 4.",
    "Final prose is deterministic rendering of verified model-selected claims; this is not free-prose generation accuracy.",
    "Gold corrections and changed delivery protocol prevent a controlled accuracy comparison with v6."
  ]
}