{
  "submission": "team8_HIPE-2026-v1.0-impresso-test-en_run1.jsonl",
  "reference": "HIPE-2026-v1.0-impresso-test-en.jsonl",
  "team": "team8",
  "dataset": "impresso",
  "split": "test",
  "language": "en",
  "run": "run1",
  "at_label_mode": "TERNARY",
  "confusion_matrices": {
    "at": {
      "labels": [
        "TRUE",
        "PROBABLE",
        "FALSE"
      ],
      "matrix": [
        [
          54,
          16,
          5
        ],
        [
          1,
          18,
          9
        ],
        [
          1,
          7,
          51
        ]
      ],
      "rows": "gold",
      "columns": "prediction",
      "total": 162,
      "table": [
        {
          "gold": "TRUE",
          "pred_TRUE": 54,
          "pred_PROBABLE": 16,
          "pred_FALSE": 5
        },
        {
          "gold": "PROBABLE",
          "pred_TRUE": 1,
          "pred_PROBABLE": 18,
          "pred_FALSE": 9
        },
        {
          "gold": "FALSE",
          "pred_TRUE": 1,
          "pred_PROBABLE": 7,
          "pred_FALSE": 51
        }
      ],
      "accuracy": 0.7593,
      "per_label": {
        "TRUE": {
          "precision": 0.9643,
          "recall": 0.72,
          "f1": 0.8244,
          "support": 75,
          "true_positive": 54,
          "false_positive": 2,
          "false_negative": 21
        },
        "PROBABLE": {
          "precision": 0.439,
          "recall": 0.6429,
          "f1": 0.5217,
          "support": 28,
          "true_positive": 18,
          "false_positive": 23,
          "false_negative": 10
        },
        "FALSE": {
          "precision": 0.7846,
          "recall": 0.8644,
          "f1": 0.8226,
          "support": 59,
          "true_positive": 51,
          "false_positive": 14,
          "false_negative": 8
        }
      }
    },
    "isAt": {
      "labels": [
        "TRUE",
        "FALSE"
      ],
      "matrix": [
        [
          38,
          27
        ],
        [
          7,
          90
        ]
      ],
      "rows": "gold",
      "columns": "prediction",
      "total": 162,
      "table": [
        {
          "gold": "TRUE",
          "pred_TRUE": 38,
          "pred_FALSE": 27
        },
        {
          "gold": "FALSE",
          "pred_TRUE": 7,
          "pred_FALSE": 90
        }
      ],
      "accuracy": 0.7901,
      "per_label": {
        "TRUE": {
          "precision": 0.8444,
          "recall": 0.5846,
          "f1": 0.6909,
          "support": 65,
          "true_positive": 38,
          "false_positive": 7,
          "false_negative": 27
        },
        "FALSE": {
          "precision": 0.7692,
          "recall": 0.9278,
          "f1": 0.8411,
          "support": 97,
          "true_positive": 90,
          "false_positive": 27,
          "false_negative": 7
        }
      }
    }
  }
}