{
  "submission": "team9_HIPE-2026-v1.0-surprise-test-fr_run2.jsonl",
  "reference": "HIPE-2026-v1.0-surprise-test-fr.jsonl",
  "team": "team9",
  "dataset": "surprise",
  "split": "test",
  "language": "fr",
  "run": "run2",
  "at_label_mode": "TERNARY",
  "confusion_matrices": {
    "at": {
      "labels": [
        "TRUE",
        "PROBABLE",
        "FALSE"
      ],
      "matrix": [
        [
          70,
          54,
          6
        ],
        [
          7,
          41,
          12
        ],
        [
          18,
          74,
          198
        ]
      ],
      "rows": "gold",
      "columns": "prediction",
      "total": 480,
      "table": [
        {
          "gold": "TRUE",
          "pred_TRUE": 70,
          "pred_PROBABLE": 54,
          "pred_FALSE": 6
        },
        {
          "gold": "PROBABLE",
          "pred_TRUE": 7,
          "pred_PROBABLE": 41,
          "pred_FALSE": 12
        },
        {
          "gold": "FALSE",
          "pred_TRUE": 18,
          "pred_PROBABLE": 74,
          "pred_FALSE": 198
        }
      ],
      "accuracy": 0.6438,
      "per_label": {
        "TRUE": {
          "precision": 0.7368,
          "recall": 0.5385,
          "f1": 0.6222,
          "support": 130,
          "true_positive": 70,
          "false_positive": 25,
          "false_negative": 60
        },
        "PROBABLE": {
          "precision": 0.2426,
          "recall": 0.6833,
          "f1": 0.3581,
          "support": 60,
          "true_positive": 41,
          "false_positive": 128,
          "false_negative": 19
        },
        "FALSE": {
          "precision": 0.9167,
          "recall": 0.6828,
          "f1": 0.7826,
          "support": 290,
          "true_positive": 198,
          "false_positive": 18,
          "false_negative": 92
        }
      }
    }
  }
}