{
  "scope": "Frozen matched final comparison; primary20percent; intervals rounded to one decimal; warnings are post-study descriptive",
  "rows": [
    {
      "dataset": "FinQA",
      "model": "GPT-5.4 mini",
      "n": 455,
      "slots": 91,
      "total": 127,
      "c": 39,
      "v": 64,
      "warned": 68,
      "flagged": 106,
      "selectedWarned": 61,
      "range": [
        14.6,
        46
      ],
      "budgets": [
        [
          12,
          18
        ],
        [
          23,
          40
        ],
        [
          39,
          64
        ],
        [
          72,
          88
        ]
      ]
    },
    {
      "dataset": "FinQA",
      "model": "GPT-6.1 Sol",
      "n": 455,
      "slots": 91,
      "total": 73,
      "c": 38,
      "v": 13,
      "warned": 2,
      "flagged": 7,
      "selectedWarned": 2,
      "range": [
        -38,
        -14.1
      ],
      "budgets": [
        [
          12,
          3
        ],
        [
          21,
          7
        ],
        [
          38,
          13
        ],
        [
          57,
          38
        ]
      ]
    },
    {
      "dataset": "FinQA",
      "model": "Claude Sonnet 5.5",
      "n": 455,
      "slots": 91,
      "total": 100,
      "c": 48,
      "v": 52,
      "warned": 47,
      "flagged": 71,
      "selectedWarned": 47,
      "range": [
        -7.5,
        17.3
      ],
      "budgets": [
        [
          11,
          17
        ],
        [
          24,
          34
        ],
        [
          48,
          52
        ],
        [
          73,
          65
        ]
      ]
    },
    {
      "dataset": "FinQA",
      "model": "Gemini 4 Argon",
      "n": 455,
      "slots": 91,
      "total": 67,
      "c": 36,
      "v": 13,
      "warned": 4,
      "flagged": 18,
      "selectedWarned": 4,
      "range": [
        -34.2,
        -14.4
      ],
      "budgets": [
        [
          11,
          4
        ],
        [
          21,
          8
        ],
        [
          36,
          13
        ],
        [
          53,
          36
        ]
      ]
    },
    {
      "dataset": "TAT-QA",
      "model": "GPT-5.4 mini",
      "n": 487,
      "slots": 97,
      "total": 88,
      "c": 24,
      "v": 46,
      "warned": 45,
      "flagged": 118,
      "selectedWarned": 43,
      "range": [
        10.3,
        35.4
      ],
      "budgets": [
        [
          12,
          21
        ],
        [
          17,
          40
        ],
        [
          24,
          46
        ],
        [
          45,
          59
        ]
      ]
    },
    {
      "dataset": "TAT-QA",
      "model": "GPT-6.1 Sol",
      "n": 487,
      "slots": 97,
      "total": 42,
      "c": 20,
      "v": 7,
      "warned": 3,
      "flagged": 26,
      "selectedWarned": 3,
      "range": [
        -22.7,
        -5
      ],
      "budgets": [
        [
          11,
          3
        ],
        [
          15,
          4
        ],
        [
          20,
          7
        ],
        [
          29,
          20
        ]
      ]
    },
    {
      "dataset": "TAT-QA",
      "model": "Claude Sonnet 5.5",
      "n": 487,
      "slots": 97,
      "total": 83,
      "c": 39,
      "v": 51,
      "warned": 47,
      "flagged": 71,
      "selectedWarned": 47,
      "range": [
        -1,
        24.7
      ],
      "budgets": [
        [
          15,
          20
        ],
        [
          26,
          44
        ],
        [
          39,
          51
        ],
        [
          59,
          61
        ]
      ]
    },
    {
      "dataset": "TAT-QA",
      "model": "Gemini 4 Argon",
      "n": 487,
      "slots": 97,
      "total": 36,
      "c": 24,
      "v": 10,
      "warned": 3,
      "flagged": 24,
      "selectedWarned": 3,
      "range": [
        -24.5,
        -8
      ],
      "budgets": [
        [
          11,
          3
        ],
        [
          16,
          3
        ],
        [
          24,
          10
        ],
        [
          29,
          18
        ]
      ]
    }
  ]
}