{
  "schema": "fidelity_mr_research_data_v1",
  "status": "PUBLISHABLE — publication boundary open at generation time",
  "publishable": true,
  "not_publishable_reason": null,
  "framing_rule": "NEVER \"X% of AI citations are supported.\" ONLY \"of the N resolvable citations for engine E, X% were supported; resolvable coverage was C%.\" AND never blend cohorts of different semantics — each semantic cohort_key is reported as its OWN track; bridges report comparability but never merge.",
  "generated_at": "2026-09-07T11:31:42.954Z",
  "aggregate_sha256": "b8c28c76f41cd7710dad00e77fc1b47e0b84a9454521a13873764660246ecfe8",
  "corpus_receipt_sha256": "29c9d69d37c5bfb55b595f2107dd8069b710a62a8b1d0a6f6a4c50cc215816a8",
  "instrument": {
    "versions": [
      {
        "kind": "corpus",
        "seq": 0,
        "instrument_version": "f9979596c1aac66a",
        "code_hash": "d5ea378808399bbb"
      },
      {
        "kind": "cycle",
        "seq": 1,
        "instrument_version": "2ab0c45cdfc06ea2",
        "code_hash": "0f3e75827dbd08bb"
      },
      {
        "kind": "cycle",
        "seq": 2,
        "instrument_version": "2ab0c45cdfc06ea2",
        "code_hash": "0f3e75827dbd08bb"
      },
      {
        "kind": "cycle",
        "seq": 3,
        "instrument_version": "2ab0c45cdfc06ea2",
        "code_hash": "0f3e75827dbd08bb"
      },
      {
        "kind": "cycle",
        "seq": 4,
        "instrument_version": "2ab0c45cdfc06ea2",
        "code_hash": "0f3e75827dbd08bb"
      },
      {
        "kind": "cycle",
        "seq": 5,
        "instrument_version": "2ab0c45cdfc06ea2",
        "code_hash": "0f3e75827dbd08bb"
      }
    ],
    "distinct_instrument_versions": [
      "f9979596c1aac66a",
      "2ab0c45cdfc06ea2"
    ],
    "heterogeneous": true,
    "note": "Instrument code versions are PROVENANCE, not a measurement boundary. Cohorts partition by SEMANTIC cohort_key (fetch basis, prompt/model/claim-map pins, label set, adjudication + denominator/evidence policy). Two code versions with identical measurement semantics pool automatically; a semantic change splits — independent of this flag. A bridge sample measures cross-cohort comparability and is REPORTED, never used to merge cohorts."
  },
  "cohort_key_spec": {
    "fields": [
      "fetch_basis",
      "prompt_hash",
      "models",
      "claim_map_version",
      "label_set",
      "adjudication_policy",
      "denominator_and_evidence_policy"
    ],
    "provenance_only": [
      "instrument_version",
      "code_hash"
    ],
    "rule": "Cohorts partition MECHANICALLY by cohort_key. Identical semantics pool automatically (the key matches) — no flag, no ack, no bridge-authorized pooling. Any semantic field differing splits the cohort. Operational code hashes are PROVENANCE and NEVER enter the key. Bridges measure comparability and are reported alongside, but NEVER merge two cohorts."
  },
  "comparability_bridges": [],
  "lineage": [
    {
      "seq": 0,
      "kind": "corpus",
      "receipt_sha256": "29c9d69d37c5bfb55b595f2107dd8069b710a62a8b1d0a6f6a4c50cc215816a8",
      "manifest_id": "fidelity_claim_map_v2+2026-07-10+6f4d069d6bb8",
      "events": 10066,
      "spend_usd": 25.613406,
      "instrument_version": "f9979596c1aac66a"
    },
    {
      "seq": 1,
      "kind": "cycle",
      "receipt_sha256": "95eee2e06aeb44436bc15e2fc4c6294016f265deeede67c1ad7f5c9360ae1ce2",
      "prev_receipt_sha256": "29c9d69d37c5bfb55b595f2107dd8069b710a62a8b1d0a6f6a4c50cc215816a8",
      "resolved_events": 348,
      "spend_usd": 2.200605,
      "instrument_version": "2ab0c45cdfc06ea2"
    },
    {
      "seq": 2,
      "kind": "cycle",
      "receipt_sha256": "0f6595a976e52a7498650d862f9c322425bb1617f9345a35f7ae7d53b23e6e97",
      "prev_receipt_sha256": "95eee2e06aeb44436bc15e2fc4c6294016f265deeede67c1ad7f5c9360ae1ce2",
      "resolved_events": 308,
      "spend_usd": 1.877768,
      "instrument_version": "2ab0c45cdfc06ea2"
    },
    {
      "seq": 3,
      "kind": "cycle",
      "receipt_sha256": "6d9bea1ed6f74de400b4ec412778e2a2e63a30b35ae73fc1479b9f71edcae46d",
      "prev_receipt_sha256": "0f6595a976e52a7498650d862f9c322425bb1617f9345a35f7ae7d53b23e6e97",
      "resolved_events": 278,
      "spend_usd": 1.638498,
      "instrument_version": "2ab0c45cdfc06ea2"
    },
    {
      "seq": 4,
      "kind": "cycle",
      "receipt_sha256": "a5acc39984a2cc5597d134a7a5588bc9c4abe4cb63ad24d40f63580a3f300132",
      "prev_receipt_sha256": "6d9bea1ed6f74de400b4ec412778e2a2e63a30b35ae73fc1479b9f71edcae46d",
      "resolved_events": 235,
      "spend_usd": 1.399401,
      "instrument_version": "2ab0c45cdfc06ea2"
    },
    {
      "seq": 5,
      "kind": "cycle",
      "receipt_sha256": "f7b04b4c870745dc8594d13bd99446d417cea02c05e0c9c248e54cf6ca2a4109",
      "prev_receipt_sha256": "a5acc39984a2cc5597d134a7a5588bc9c4abe4cb63ad24d40f63580a3f300132",
      "resolved_events": 206,
      "spend_usd": 1.084854,
      "instrument_version": "2ab0c45cdfc06ea2"
    }
  ],
  "cumulative": {
    "total_denominator_events": 11441,
    "outcomes": {
      "graded": 4173,
      "unmappable": 6017,
      "unreadable": 769,
      "uncertain": 482
    },
    "cohort_arithmetic": {
      "events_by_cohort": {
        "legacy_baseline": 10066,
        "collection_time_series__bounded_retry_defer": 1375
      },
      "outcomes_by_cohort": {
        "legacy_baseline": {
          "graded": 3033,
          "unmappable": 6017,
          "unreadable": 666,
          "uncertain": 350
        },
        "collection_time_series__bounded_retry_defer": {
          "graded": 1140,
          "unmappable": 0,
          "unreadable": 103,
          "uncertain": 132
        }
      }
    },
    "spend_usd": 33.814532,
    "spend_calls": 12456
  },
  "calibration": {
    "refset_sha256": "efa6a0ae92416fa462ba70f5e53652e067353dc7d78be557e32d68db58df1a2c",
    "refset_version": "fidelity_reference_set_v3",
    "inter_model_kappa": 0.647,
    "inter_model_raw_agreement": 0.722,
    "control_accuracy": 1,
    "control_n": 30,
    "certificate": {
      "schema": "fidelity_calibration_certificate_v1",
      "certificate_sha256": "7ff08eeaf58518034cfec0b0cd447ffcab591daf88b26b98610786549dac6585",
      "instrument_sha256": "0a1fac228ad9a2733775d25638fb53bea5bbaadf1d0be02c6c8bf7ef3e5a04c8",
      "evaluation_manifest_sha256": "fe6ba53ec479bca3090718f851c3be93a5abba61971a3c7e40df848881e83b78",
      "policy_version": "fidelity_binary_support_publication_policy_v1",
      "metrics": {
        "n_scored": 91,
        "correct": 83,
        "n_excluded": 6,
        "accuracy": 0.9120879120879121,
        "wilson_lower_bound": 0.8360136111636226,
        "per_class": {
          "supported": {
            "support": 29,
            "correct": 27,
            "accuracy": 0.9310344827586207
          },
          "unsupported": {
            "support": 62,
            "correct": 56,
            "accuracy": 0.9032258064516129
          }
        }
      }
    },
    "publishable": true,
    "not_publishable_reason": null,
    "primary_excluded_holdout": {
      "agreement": null,
      "n": 0
    }
  },
  "benchmark": {
    "source": "AttrScore (AttrEval-GenSearch)",
    "source_url": "https://huggingface.co/datasets/osunlp/AttrScore",
    "dataset_revision": "467dcdd2cd31f9b5e8625491f3bdf7af90943a8d",
    "source_sha256": "8f8470404b92ce4a99e2d9fa6685fd0805afac877a86291bdacbf44b0885fb93",
    "instrument_sha256": "0a1fac228ad9a2733775d25638fb53bea5bbaadf1d0be02c6c8bf7ef3e5a04c8",
    "validated_at": "2026-07-15T05:26:19.174Z",
    "public_metric": "binary_claim_support_agreement",
    "agreement_vs_benchmark": 0.9166666666666666,
    "wilson_lower_bound": 0.8193105798166558,
    "n_scored": 60,
    "binary": {
      "n_scored": 60,
      "correct": 55,
      "n_excluded": 6,
      "accuracy": 0.9166666666666666,
      "wilson_lower_bound": 0.8193105798166558,
      "per_class": {
        "supported": {
          "support": 22,
          "correct": 19,
          "accuracy": 0.8636363636363636
        },
        "unsupported": {
          "support": 38,
          "correct": 36,
          "accuracy": 0.9473684210526315
        }
      }
    },
    "subtype_metrics_role": "diagnostic-only; never publication eligibility",
    "per_label": {
      "attributable": {
        "support": 22,
        "agree": 19,
        "agreement": 0.864
      },
      "not_attributable": {
        "support": 20,
        "agree": 19,
        "agreement": 0.95
      },
      "extrapolatory": {
        "support": 18,
        "agree": 6,
        "agreement": 0.333
      }
    },
    "extrapolatory_confusion": {
      "support": 18,
      "to_not_attributable": 11,
      "to_attributable": 1,
      "to_extrapolatory": 6
    }
  },
  "cohorts": {
    "legacy_baseline": {
      "cohort_key": "c74383df31771cfe",
      "cohort_key_fields": {
        "fetch_basis": "grade-time",
        "prompt_hash": "acaee3b4b10e6a26",
        "models": {
          "grader_a": "gemini-2.5-flash",
          "grader_b": "gpt-4.1-mini-2025-04-14",
          "judge": "claude-sonnet-5"
        },
        "claim_map_version": "fidelity_claim_map_v2",
        "label_set": "fidelity-5label-v1(Supported|Amplified|Contradicted|Misattributed|Fabricated)",
        "adjudication_policy": "always-dual",
        "denominator_and_evidence_policy": "grade-time-fetch"
      },
      "label": "Legacy baseline — frozen corpus, grade-time fetch basis",
      "fetch_time_basis": "grade-time",
      "denominator_evidence_policy": "grade-time-fetch",
      "provenance": [
        {
          "kind": "corpus",
          "seq": 0,
          "instrument_version": "f9979596c1aac66a",
          "code_hash": "d5ea378808399bbb",
          "events": 10066
        }
      ],
      "events": 10066,
      "overall": {
        "status": "reported",
        "coverage": 0.3013,
        "statement": "Of the 3033 resolvable citations for this engine, 49.7% were Supported; resolvable coverage was 30.1% (3033/10066).",
        "supported_rate_over_resolved": 0.4972
      },
      "by_engine": {
        "chatgpt": {
          "status": "reported",
          "coverage": 0.7699,
          "statement": "Of the 1255 resolvable citations for this engine, 54.8% were Supported; resolvable coverage was 77% (1255/1630).",
          "supported_rate_over_resolved": 0.5482
        },
        "claude": {
          "status": "collecting",
          "coverage": 0.0647,
          "statement": "Engine still COLLECTING: only 88 of 1361 citations were resolvable (coverage 6.5%) — below the evidence floor, so no fidelity rate is reported.",
          "supported_rate_over_resolved": null
        },
        "gemini": {
          "status": "collecting",
          "coverage": 0.0901,
          "statement": "Engine still COLLECTING: only 392 of 4349 citations were resolvable (coverage 9%) — below the evidence floor, so no fidelity rate is reported.",
          "supported_rate_over_resolved": null
        },
        "perplexity": {
          "status": "reported",
          "coverage": 0.4762,
          "statement": "Of the 1298 resolvable citations for this engine, 50.1% were Supported; resolvable coverage was 47.6% (1298/2726).",
          "supported_rate_over_resolved": 0.5008
        }
      }
    },
    "collection_time_series__bounded_retry_defer": {
      "cohort_key": "06ddb22c15fa05d0",
      "cohort_key_fields": {
        "fetch_basis": "collection-time-snapshot",
        "prompt_hash": "acaee3b4b10e6a26",
        "models": {
          "grader_a": "gemini-2.5-flash",
          "grader_b": "gpt-4.1-mini-2025-04-14",
          "judge": "claude-sonnet-5"
        },
        "claim_map_version": "fidelity_claim_map_v2",
        "label_set": "fidelity-5label-v1(Supported|Amplified|Contradicted|Misattributed|Fabricated)",
        "adjudication_policy": "always-dual",
        "denominator_and_evidence_policy": "collection-snapshot+bounded-retry-defer"
      },
      "label": "Collection-time series — snapshot basis, bounded snapshot retry + deferral (F4-3)",
      "fetch_time_basis": "collection-time snapshot",
      "denominator_evidence_policy": "collection-snapshot+bounded-retry-defer",
      "provenance": [
        {
          "kind": "cycle",
          "seq": 1,
          "instrument_version": "2ab0c45cdfc06ea2",
          "code_hash": "0f3e75827dbd08bb",
          "events": 348
        },
        {
          "kind": "cycle",
          "seq": 2,
          "instrument_version": "2ab0c45cdfc06ea2",
          "code_hash": "0f3e75827dbd08bb",
          "events": 308
        },
        {
          "kind": "cycle",
          "seq": 3,
          "instrument_version": "2ab0c45cdfc06ea2",
          "code_hash": "0f3e75827dbd08bb",
          "events": 278
        },
        {
          "kind": "cycle",
          "seq": 4,
          "instrument_version": "2ab0c45cdfc06ea2",
          "code_hash": "0f3e75827dbd08bb",
          "events": 235
        },
        {
          "kind": "cycle",
          "seq": 5,
          "instrument_version": "2ab0c45cdfc06ea2",
          "code_hash": "0f3e75827dbd08bb",
          "events": 206
        }
      ],
      "events": 1375,
      "overall": {
        "status": "reported",
        "coverage": 0.8291,
        "statement": "Of the 1140 resolvable citations for this engine, 51.6% were Supported; resolvable coverage was 82.9% (1140/1375).",
        "supported_rate_over_resolved": 0.5158
      },
      "by_engine": {
        "chatgpt": {
          "status": "reported",
          "coverage": 0.8291,
          "statement": "Of the 1140 resolvable citations for this engine, 51.6% were Supported; resolvable coverage was 82.9% (1140/1375).",
          "supported_rate_over_resolved": 0.5158
        },
        "claude": {
          "status": "collecting",
          "coverage": null,
          "statement": "Engine still COLLECTING: only 0 of 0 citations were resolvable (coverage 0%) — below the evidence floor, so no fidelity rate is reported.",
          "supported_rate_over_resolved": null
        },
        "gemini": {
          "status": "collecting",
          "coverage": null,
          "statement": "Engine still COLLECTING: only 0 of 0 citations were resolvable (coverage 0%) — below the evidence floor, so no fidelity rate is reported.",
          "supported_rate_over_resolved": null
        },
        "perplexity": {
          "status": "collecting",
          "coverage": null,
          "statement": "Engine still COLLECTING: only 0 of 0 citations were resolvable (coverage 0%) — below the evidence floor, so no fidelity rate is reported.",
          "supported_rate_over_resolved": null
        }
      }
    }
  },
  "cycle_series": [
    {
      "cycle_seq": 1,
      "instrument_version": "2ab0c45cdfc06ea2",
      "cycle_receipt_sha256": "95eee2e06aeb44436bc15e2fc4c6294016f265deeede67c1ad7f5c9360ae1ce2",
      "prev_receipt_sha256": "29c9d69d37c5bfb55b595f2107dd8069b710a62a8b1d0a6f6a4c50cc215816a8",
      "graded": 309,
      "unmappable": 0,
      "unreadable": 14,
      "uncertain": 25,
      "this_cycle_spend_usd": 2.200605,
      "aggregate_sha256": "0cd286ed31382ff5c24b82b7e629e68b29787b1991a18996cf855cd5f26234fd",
      "at": "2026-09-02T22:21:49.367Z"
    },
    {
      "cycle_seq": 2,
      "instrument_version": "2ab0c45cdfc06ea2",
      "cycle_receipt_sha256": "0f6595a976e52a7498650d862f9c322425bb1617f9345a35f7ae7d53b23e6e97",
      "prev_receipt_sha256": "95eee2e06aeb44436bc15e2fc4c6294016f265deeede67c1ad7f5c9360ae1ce2",
      "graded": 257,
      "unmappable": 0,
      "unreadable": 17,
      "uncertain": 34,
      "this_cycle_spend_usd": 1.877768,
      "aggregate_sha256": "55936204cf4880077f301a366723240e0b03a96858962c31987a209d080bc0c2",
      "at": "2026-09-03T11:35:30.542Z"
    },
    {
      "cycle_seq": 3,
      "instrument_version": "2ab0c45cdfc06ea2",
      "cycle_receipt_sha256": "6d9bea1ed6f74de400b4ec412778e2a2e63a30b35ae73fc1479b9f71edcae46d",
      "prev_receipt_sha256": "0f6595a976e52a7498650d862f9c322425bb1617f9345a35f7ae7d53b23e6e97",
      "graded": 227,
      "unmappable": 0,
      "unreadable": 22,
      "uncertain": 29,
      "this_cycle_spend_usd": 1.638498,
      "aggregate_sha256": "4c5c7260fba7dcc1b5a1b677cae06423a0d1266ab31c44d5baa5c18873a4887b",
      "at": "2026-09-04T11:46:17.331Z"
    },
    {
      "cycle_seq": 4,
      "instrument_version": "2ab0c45cdfc06ea2",
      "cycle_receipt_sha256": "a5acc39984a2cc5597d134a7a5588bc9c4abe4cb63ad24d40f63580a3f300132",
      "prev_receipt_sha256": "6d9bea1ed6f74de400b4ec412778e2a2e63a30b35ae73fc1479b9f71edcae46d",
      "graded": 185,
      "unmappable": 0,
      "unreadable": 24,
      "uncertain": 26,
      "this_cycle_spend_usd": 1.399401,
      "aggregate_sha256": "625a7e05727ee16035abe9263dcad3ba1af94d06dee859e1d675e1c59f6d1837",
      "at": "2026-09-05T11:34:02.404Z"
    },
    {
      "cycle_seq": 5,
      "instrument_version": "2ab0c45cdfc06ea2",
      "cycle_receipt_sha256": "f7b04b4c870745dc8594d13bd99446d417cea02c05e0c9c248e54cf6ca2a4109",
      "prev_receipt_sha256": "a5acc39984a2cc5597d134a7a5588bc9c4abe4cb63ad24d40f63580a3f300132",
      "graded": 162,
      "unmappable": 0,
      "unreadable": 26,
      "uncertain": 18,
      "this_cycle_spend_usd": 1.084854,
      "aggregate_sha256": "b8c28c76f41cd7710dad00e77fc1b47e0b84a9454521a13873764660246ecfe8",
      "at": "2026-09-06T11:32:55.179Z"
    }
  ]
}