{
  "submission": "team14_HIPE-2026-v1.0-surprise-test-fr_run1.jsonl",
  "reference": "HIPE-2026-v1.0-surprise-test-fr.jsonl",
  "team": "team14",
  "dataset": "surprise",
  "split": "test",
  "language": "fr",
  "run": "run1",
  "at_label_mode": "TERNARY",
  "confusion_matrices": {
    "at": {
      "labels": [
        "TRUE",
        "PROBABLE",
        "FALSE"
      ],
      "matrix": [
        [
          74,
          1,
          55
        ],
        [
          29,
          0,
          31
        ],
        [
          48,
          0,
          242
        ]
      ],
      "rows": "gold",
      "columns": "prediction",
      "total": 480,
      "table": [
        {
          "gold": "TRUE",
          "pred_TRUE": 74,
          "pred_PROBABLE": 1,
          "pred_FALSE": 55
        },
        {
          "gold": "PROBABLE",
          "pred_TRUE": 29,
          "pred_PROBABLE": 0,
          "pred_FALSE": 31
        },
        {
          "gold": "FALSE",
          "pred_TRUE": 48,
          "pred_PROBABLE": 0,
          "pred_FALSE": 242
        }
      ],
      "accuracy": 0.6583,
      "per_label": {
        "TRUE": {
          "precision": 0.4901,
          "recall": 0.5692,
          "f1": 0.5267,
          "support": 130,
          "true_positive": 74,
          "false_positive": 77,
          "false_negative": 56
        },
        "PROBABLE": {
          "precision": 0.0,
          "recall": 0.0,
          "f1": 0.0,
          "support": 60,
          "true_positive": 0,
          "false_positive": 1,
          "false_negative": 60
        },
        "FALSE": {
          "precision": 0.7378,
          "recall": 0.8345,
          "f1": 0.7832,
          "support": 290,
          "true_positive": 242,
          "false_positive": 86,
          "false_negative": 48
        }
      }
    }
  }
}