{
  "submission": "team11_HIPE-2026-v1.0-impresso-test-en_run2.jsonl",
  "reference": "HIPE-2026-v1.0-impresso-test-en.jsonl",
  "team": "team11",
  "dataset": "impresso",
  "split": "test",
  "language": "en",
  "run": "run2",
  "at_label_mode": "TERNARY",
  "confusion_matrices": {
    "at": {
      "labels": [
        "TRUE",
        "PROBABLE",
        "FALSE"
      ],
      "matrix": [
        [
          27,
          42,
          6
        ],
        [
          1,
          20,
          7
        ],
        [
          1,
          19,
          39
        ]
      ],
      "rows": "gold",
      "columns": "prediction",
      "total": 162,
      "table": [
        {
          "gold": "TRUE",
          "pred_TRUE": 27,
          "pred_PROBABLE": 42,
          "pred_FALSE": 6
        },
        {
          "gold": "PROBABLE",
          "pred_TRUE": 1,
          "pred_PROBABLE": 20,
          "pred_FALSE": 7
        },
        {
          "gold": "FALSE",
          "pred_TRUE": 1,
          "pred_PROBABLE": 19,
          "pred_FALSE": 39
        }
      ],
      "accuracy": 0.5309,
      "per_label": {
        "TRUE": {
          "precision": 0.931,
          "recall": 0.36,
          "f1": 0.5192,
          "support": 75,
          "true_positive": 27,
          "false_positive": 2,
          "false_negative": 48
        },
        "PROBABLE": {
          "precision": 0.2469,
          "recall": 0.7143,
          "f1": 0.367,
          "support": 28,
          "true_positive": 20,
          "false_positive": 61,
          "false_negative": 8
        },
        "FALSE": {
          "precision": 0.75,
          "recall": 0.661,
          "f1": 0.7027,
          "support": 59,
          "true_positive": 39,
          "false_positive": 13,
          "false_negative": 20
        }
      }
    },
    "isAt": {
      "labels": [
        "TRUE",
        "FALSE"
      ],
      "matrix": [
        [
          21,
          44
        ],
        [
          1,
          96
        ]
      ],
      "rows": "gold",
      "columns": "prediction",
      "total": 162,
      "table": [
        {
          "gold": "TRUE",
          "pred_TRUE": 21,
          "pred_FALSE": 44
        },
        {
          "gold": "FALSE",
          "pred_TRUE": 1,
          "pred_FALSE": 96
        }
      ],
      "accuracy": 0.7222,
      "per_label": {
        "TRUE": {
          "precision": 0.9545,
          "recall": 0.3231,
          "f1": 0.4828,
          "support": 65,
          "true_positive": 21,
          "false_positive": 1,
          "false_negative": 44
        },
        "FALSE": {
          "precision": 0.6857,
          "recall": 0.9897,
          "f1": 0.8101,
          "support": 97,
          "true_positive": 96,
          "false_positive": 44,
          "false_negative": 1
        }
      }
    }
  }
}