{
  "reviews": [
    {
      "task": "quac/answerability",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "coarse source/recast inspection",
      "verdict": "retain_with_source_caveat",
      "sampled_examples": [
        {
          "id": "C_324c6c1e2e9d4d4984218d84afacf704_1_q#8",
          "split": "train",
          "recast_sha256": "43808c6eac2bdaf4f589768316aa0f92dea2bf4a1982b70bc901277093b432d1"
        }
      ],
      "findings": "The inspected question about tour companions is answerable from the complete passage. Earlier teacher answers remain source-native and can be noisy.",
      "provenance": {
        "dataset": "allenai/quac",
        "config": "plain_text",
        "revision": "768addbf2099da8efabd67d3a669f6483b813c1b"
      },
      "limitations": "Small sample; not an estimate of corpus-wide label accuracy."
    },
    {
      "task": "hatexplain/majority",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "coarse source/recast inspection",
      "verdict": "retain_with_source_caveat",
      "sampled_examples": [
        {
          "id": "1179082934357118976_twitter",
          "split": "train",
          "recast_sha256": "d49ad7c8cfcce14b3f9f404c58bd3054afcb55e4e8ff4c4f933255e7308cdfbc"
        }
      ],
      "findings": "Majority votes are preserved, ties excluded. The inspected abusive narrative was labeled normal by source annotators: labels concern hate/offense judgments, not all disturbing content.",
      "provenance": {
        "dataset": "Hate-speech-CNERG/hatexplain",
        "revision": "b0e5676acc1d0dcb31d6ce8e93e7e94109044ed6"
      },
      "limitations": "Small sample; not an estimate of corpus-wide label accuracy."
    },
    {
      "task": "hotpotqa/supporting-fact",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "coarse source/recast inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "5a890ba1554299669944a47f:Eastern Hills Mall:0",
          "split": "train",
          "recast_sha256": "67c57cea8d7ad643e1f5c354f298274b08fff5dbc7e6362f60d1974a056ac280"
        }
      ],
      "findings": "The inspected distractor sentence is not a native supporting fact. Membership uses question/title/sentence identity, not text matching.",
      "provenance": {
        "dataset": "hotpotqa/hotpot_qa",
        "config": "distractor",
        "revision": "1908d6afbbead072334abe2965f91bd2709910ab"
      },
      "limitations": "Small sample; not an estimate of corpus-wide label accuracy."
    },
    {
      "task": "multiwoz22/dialogue-act",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "coarse source/recast inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "PMUL3854.json:10",
          "split": "train",
          "recast_sha256": "6f6003cf657c3c1d0e39bfad284caa52b2a34c0d5a07ccfd58ac9c94487ad19e"
        }
      ],
      "findings": "The inspected Hotel-Inform turn supplies booking parameters. Only rows with one unique native dialogue-act type are emitted.",
      "provenance": {
        "dataset": "pfb30/multi_woz_v22",
        "config": "v2.2",
        "revision": "a65d2d3261c42a309b634179c186628bf35d9298"
      },
      "limitations": "Small sample; not an estimate of corpus-wide label accuracy."
    },
    {
      "task": "multiwoz22/categorical-slot",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "coarse source/recast inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "PMUL3922.json:2:hotel-stars",
          "split": "train",
          "recast_sha256": "a7c0849bb28b8e2869516d88ff93440ba1a2c364290ba57efec614ddbe3371b5"
        }
      ],
      "findings": "The inspected hotel-stars=3 value is grounded in the earlier user request; current state is not shown in inputs.",
      "provenance": {
        "dataset": "pfb30/multi_woz_v22",
        "config": "v2.2",
        "revision": "a65d2d3261c42a309b634179c186628bf35d9298"
      },
      "limitations": "Small sample; not an estimate of corpus-wide label accuracy."
    },
    {
      "task": "schema-guided/intent",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "coarse source/recast inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "99_00115:2:Media_1",
          "split": "train",
          "recast_sha256": "ac910b0d66e94147cbdfca658adb6d7011b89b8d46aa41f7a6bc96b7310ccf62"
        }
      ],
      "findings": "The inspected Media_1 intent remains FindMovies while another service is introduced. Instructions ask for the named service state, not a single global utterance intent.",
      "provenance": {
        "dataset": "google-research-datasets/schema_guided_dstc8",
        "config": "dialogues",
        "revision": "a8029ad8c12e13cdbfefc2f649e245fa135bc98a"
      },
      "limitations": "Small sample; not an estimate of corpus-wide label accuracy."
    },
    {
      "task": "coqa/yesno",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "coarse source/recast inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "df8754b3ef108c1950e9c4bef451f5c2232f3b59a8e29546032c4ecb153b72b0:3",
          "split": "train",
          "recast_sha256": "281d6da4ef27054297f703051b1177f18acf2fad82e26dd2a260b437afc5f081"
        }
      ],
      "findings": "The inspected no answer about wealth is supported by the passage describing Guizhou as one of the poorest provinces. Only exact native binary answers are retained.",
      "provenance": {
        "dataset": "stanfordnlp/coqa",
        "revision": "0d9e9952f1ef6e5415492d3d84b5873259137e3c"
      },
      "limitations": "Small sample; not an estimate of corpus-wide label accuracy."
    },
    {
      "task": "qasper/yesno",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "coarse source/recast inspection",
      "verdict": "retain_with_source_caveat",
      "sampled_examples": [
        {
          "id": "05238d1fad2128403577822aa4822ef8ca9570ac",
          "split": "train",
          "recast_sha256": "d8c267b9a1e02c758057e1af2018d4b100f30966df540d3bfbf6cc696b01ec8c"
        },
        {
          "id": "2a1069ae3629ae8ecc19d2305f23445c0231dc39",
          "split": "train",
          "recast_sha256": "a24d5a5f7959d01ae32fd1aea57c785f2e5adea5950f3304e72d5794b368d497"
        },
        {
          "id": "061682beb3dbd7c76cfa26f7ae650e548503d977",
          "split": "train",
          "recast_sha256": "d288739434c145be019a651a90e9d46d333720995059f2cb8a352262dc2fd00d"
        }
      ],
      "findings": "Three inspected binary cases retain text evidence and consistent native annotations. Computation-time and non-neural WSD cases are supported; a rationale-timing question is less explicit in its cited evidence and remains a source-quality caveat.",
      "provenance": {
        "dataset": "allenai/qasper",
        "config": "qasper",
        "revision": "06806e4608976fc2fac0a090ac425d5b2b29caf4"
      },
      "limitations": "Small sample; not an estimate of corpus-wide label accuracy."
    },
    {
      "task": "topv2/intent",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "coarse source/recast inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "c64fdd72d7b496fbe227586eb22d051df615cdf2ce5653ed5d2415135757b138",
          "split": "train",
          "recast_sha256": "45a5622f351fc2392caa24ebc5efa4acf95b7221e85c8241ab087e595325caea"
        }
      ],
      "findings": "The inspected remaining-time question maps to the native get_timer root intent. Nested slots and gold parse are excluded from inputs.",
      "provenance": {
        "dataset": "WillHeld/top_v2",
        "revision": "5e382a8497d4dd28842cc0bfa85387f965ac9d8d"
      },
      "limitations": "Small sample; not an estimate of corpus-wide label accuracy."
    },
    {
      "task": "slurp/intent",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "coarse source/recast inspection",
      "verdict": "retain_with_source_caveat",
      "sampled_examples": [
        {
          "id": "9900",
          "split": "train",
          "recast_sha256": "cfeb385ccb1238b4030ea3ae548800894d73fe382841a6785d1147759f76a345"
        }
      ],
      "findings": "The inspected resume-Battle-Royale utterance has native play_audiobook intent, but text alone does not explicitly distinguish media type. Native intent supervision has underspecified cases despite duplicate/short-transcript filtering.",
      "provenance": {
        "dataset": "tasksource/slurp",
        "revision": "2e033e72af1d7df94523d5053b31a30130b7f877"
      },
      "limitations": "Small sample; not an estimate of corpus-wide label accuracy."
    },
    {
      "task": "mcscript2",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "coarse source/recast inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "1044:8",
          "split": "train",
          "recast_sha256": "99df8ad69a57a48105de627fe31c9208e62945da616e5bd843482b8ee820f2ed"
        }
      ],
      "findings": "The inspected story explicitly names yellow daisies, matching the gold native option. XML choice order and gold index are retained before Jev permutation.",
      "provenance": {
        "dataset": "tasksource/mcscript2",
        "revision": "3b9c261f28c82ba15e6073990c0d70f05db60c4a"
      },
      "limitations": "Small sample; not an estimate of corpus-wide label accuracy."
    },
    {
      "task": "quac/answerability",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "exact staged export inspection",
      "verdict": "retain_with_source_caveat",
      "sampled_examples": [
        {
          "id": "quac:train:C_04a01eb83156458e9258f9a152d9618f_0_q#5:quac-answerability-bb7c4463e8",
          "split": "train",
          "recast_sha256": "2061da69184f4d57274be3c8c85fe3fb9d5938093693b2d4885ec0cec8cedd25",
          "gold": [
            "yes"
          ]
        }
      ],
      "findings": "The passage names Alix Bailey as the wife, supporting answerability. Prior native QA turns contain noisy answers, retained as conversation context.",
      "provenance": {
        "dataset": "allenai/quac",
        "config": "plain_text",
        "revision": "768addbf2099da8efabd67d3a669f6483b813c1b"
      },
      "limitations": "One exported training row inspected; not a corpus-wide accuracy estimate."
    },
    {
      "task": "hatexplain/majority",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "exact staged export inspection",
      "verdict": "retain_with_source_caveat",
      "sampled_examples": [
        {
          "id": "hatexplain:train:3826117_gab:hatexplain-majority-4b2a472634",
          "split": "train",
          "recast_sha256": "00d35820a85f7e2cdcb92879eba4f67efcfeb8af400e70702bcf195c337346b9",
          "gold": [
            "hate speech"
          ]
        }
      ],
      "findings": "The recast retains the native hate-speech majority for the inspected group-directed post. Source annotator judgments are subjective; no replacement labels were inferred.",
      "provenance": {
        "dataset": "Hate-speech-CNERG/hatexplain",
        "revision": "b0e5676acc1d0dcb31d6ce8e93e7e94109044ed6"
      },
      "limitations": "One exported training row inspected; not a corpus-wide accuracy estimate."
    },
    {
      "task": "hotpotqa/supporting-fact",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "exact staged export inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "hotpotqa:train:5ae1fb09554299234fd04382:The Formula (1980 film):2:hotpotqa-supporting-fact-274a376cc2",
          "split": "train",
          "recast_sha256": "d03a3caf85c1d53fae204cc83f1aa83ccb82cab23be12790599a44d993dacb5a",
          "gold": [
            "not a supporting fact"
          ]
        }
      ],
      "findings": "The inspected Craig T. Nelson sentence is not in the native supporting-fact set. The criterion explicitly asks for annotation membership; it does not claim all other sentences are irrelevant.",
      "provenance": {
        "dataset": "hotpotqa/hotpot_qa",
        "config": "distractor",
        "revision": "1908d6afbbead072334abe2965f91bd2709910ab"
      },
      "limitations": "One exported training row inspected; not a corpus-wide accuracy estimate."
    },
    {
      "task": "multiwoz22/dialogue-act",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "exact staged export inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "multiwoz22:train:SNG0065.json:0:multiwoz22-dialogue-act-7a1cf5f5fd",
          "split": "train",
          "recast_sha256": "0e52758a129cf4c6d92aaba61d52babbee72ada663a6270033b430565bc3e688",
          "gold": [
            "Taxi-Inform"
          ]
        }
      ],
      "findings": "The current user gives taxi pickup/destination details; native Taxi-Inform is consistent. No future response or gold state enters input.",
      "provenance": {
        "dataset": "pfb30/multi_woz_v22",
        "config": "v2.2",
        "revision": "a65d2d3261c42a309b634179c186628bf35d9298"
      },
      "limitations": "One exported training row inspected; not a corpus-wide accuracy estimate."
    },
    {
      "task": "multiwoz22/categorical-slot",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "exact staged export inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "multiwoz22:train:PMUL0242.json:18:restaurant-area:multiwoz22-categorical-slot-2a476f8ce2",
          "split": "train",
          "recast_sha256": "3abdbcbbd83b989795f2177ccf5ebc9f9475d5e1998084db377a13374f5ae00a",
          "gold": [
            "centre"
          ]
        }
      ],
      "findings": "Restaurant-area=centre is explicitly stated earlier and carried forward in the dialogue state. The question asks for current state, not a new fact in the final goodbye.",
      "provenance": {
        "dataset": "pfb30/multi_woz_v22",
        "config": "v2.2",
        "revision": "a65d2d3261c42a309b634179c186628bf35d9298"
      },
      "limitations": "One exported training row inspected; not a corpus-wide accuracy estimate."
    },
    {
      "task": "schema-guided/intent",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "exact staged export inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "schema-guided:train:24_00103:2:RentalCars_2:schema-guided-intent-3c604d923b",
          "split": "train",
          "recast_sha256": "8f933087eef09b464fcbdee9bfcf9eca689618cfccbb16d493f24f23c049ba6d",
          "gold": [
            "GetCarsAvailable"
          ]
        }
      ],
      "findings": "Sydney rental-car and March 3 context supports native GetCarsAvailable. The service is identified and only previous turns are shown.",
      "provenance": {
        "dataset": "google-research-datasets/schema_guided_dstc8",
        "config": "dialogues",
        "revision": "a8029ad8c12e13cdbfefc2f649e245fa135bc98a"
      },
      "limitations": "One exported training row inspected; not a corpus-wide accuracy estimate."
    },
    {
      "task": "coqa/yesno",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "exact staged export inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "coqa:train:5a5b3edd97c4d2ca071f0c67a98cf97ff717c08b3183cf9af389503b548ae7fe:11:coqa-yesno-6b0f120687",
          "split": "train",
          "recast_sha256": "c1c9e8832401e6a18e7610695c96734fbbf26550ed79e7b1f3049c0a29ffa12b",
          "gold": [
            "yes"
          ]
        }
      ],
      "findings": "The earlier answer lists clean living spaces; the passage lists vaccinations, food and clothing too, supporting yes to anything else.",
      "provenance": {
        "dataset": "stanfordnlp/coqa",
        "revision": "0d9e9952f1ef6e5415492d3d84b5873259137e3c"
      },
      "limitations": "One exported training row inspected; not a corpus-wide accuracy estimate."
    },
    {
      "task": "qasper/yesno",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "exact staged export inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "qasper:train:cd1034c183edf630018f47ff70b48d74d2bb1649:qasper-yesno-283c368943",
          "split": "train",
          "recast_sha256": "7a4c33ad8589bfc2135273837d41dbd2606f20169df84386919152217b2f169d",
          "gold": [
            "yes"
          ]
        }
      ],
      "findings": "The supplied paper contrasts near-random human detection with up to 97% automatic detection, supporting yes. Tightened the filter after a prior macro-F1 example depended on omitted table evidence: all cited evidence must now occur in supplied text, not merely one paragraph.",
      "provenance": {
        "dataset": "allenai/qasper",
        "config": "qasper",
        "revision": "06806e4608976fc2fac0a090ac425d5b2b29caf4"
      },
      "limitations": "One exported training row inspected; not a corpus-wide accuracy estimate."
    },
    {
      "task": "topv2/intent",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "exact staged export inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "topv2:train:54b52eeff568a74553d96b4705d6e7032a2128db9075094b9a3f7192a04e020c:topv2-intent-2b01248592",
          "split": "train",
          "recast_sha256": "21426263d901cb7f3dfc4ab8cecfd7f8da2989892a1b3b398c2512bd02e3e8c1",
          "gold": [
            "get reminder"
          ]
        }
      ],
      "findings": "The request asks to retrieve upcoming email reminders, consistent with the native get_reminder root intent. Gold semantic parse is excluded.",
      "provenance": {
        "dataset": "WillHeld/top_v2",
        "revision": "5e382a8497d4dd28842cc0bfa85387f965ac9d8d"
      },
      "limitations": "One exported training row inspected; not a corpus-wide accuracy estimate."
    },
    {
      "task": "slurp/intent",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "exact staged export inspection",
      "verdict": "retain_with_source_caveat",
      "sampled_examples": [
        {
          "id": "slurp:train:10987:slurp-intent-dd2971bc8b",
          "split": "train",
          "recast_sha256": "99f6ed4308d0e70593fe3be38eecd44cb060fc9f31dd39d0535f74b44808706b",
          "gold": [
            "lists_remove"
          ]
        }
      ],
      "findings": "Deleting a to-do list matches lists_remove. Rebuilt the mirror using scenario/action pairs (60 intents); raw aliases such as query previously conflated intents. Native raw labels remain in metadata. Some utterances still underspecify media type.",
      "provenance": {
        "dataset": "tasksource/slurp",
        "revision": "a21c5fbc1b0821a04f0ee9cb6de927f96dbd5338"
      },
      "limitations": "One exported training row inspected; not a corpus-wide accuracy estimate."
    },
    {
      "task": "mcscript2",
      "reviewer": {
        "kind": "model",
        "model": "gpt-6.1-sol",
        "model_id_source": "user-provided"
      },
      "checked_on": "2026-10-11",
      "scope": "exact staged export inspection",
      "verdict": "retain",
      "sampled_examples": [
        {
          "id": "mcscript2:train:2435:2:mcscript2-1af6d87229",
          "split": "train",
          "recast_sha256": "61291e6d79443c561d5eb33abf14ea0622a4dfa8f3c02dc9796be05f1ef4ca10",
          "gold": [
            "After they finished the job"
          ]
        }
      ],
      "findings": "Moving furniture back after finishing wallpapering is consistent with the story and native option. This is implicit script reasoning rather than a verbatim answer span.",
      "provenance": {
        "dataset": "tasksource/mcscript2",
        "revision": "3b9c261f28c82ba15e6073990c0d70f05db60c4a"
      },
      "limitations": "One exported training row inspected; not a corpus-wide accuracy estimate."
    }
  ],
  "export": {
    "configs": [
      "default",
      "full"
    ],
    "splits": {
      "train": 10272,
      "validation": 1060,
      "test": 795
    },
    "manifest": "additions/aebc640a1ba7/manifest.json"
  }
}