{
  "submission": "team10_HIPE-2026-v1.0-impresso-test-de_run3.jsonl",
  "reference": "HIPE-2026-v1.0-impresso-test-de.jsonl",
  "team": "team10",
  "dataset": "impresso",
  "split": "test",
  "language": "de",
  "run": "run3",
  "at_label_mode": "BINARY",
  "confusion_matrices": {
    "at": {
      "labels": [
        "TRUE",
        "FALSE"
      ],
      "matrix": [
        [
          38,
          54
        ],
        [
          22,
          124
        ]
      ],
      "rows": "gold",
      "columns": "prediction",
      "total": 238,
      "table": [
        {
          "gold": "TRUE",
          "pred_TRUE": 38,
          "pred_FALSE": 54
        },
        {
          "gold": "FALSE",
          "pred_TRUE": 22,
          "pred_FALSE": 124
        }
      ],
      "accuracy": 0.6807,
      "per_label": {
        "TRUE": {
          "precision": 0.6333,
          "recall": 0.413,
          "f1": 0.5,
          "support": 92,
          "true_positive": 38,
          "false_positive": 22,
          "false_negative": 54
        },
        "FALSE": {
          "precision": 0.6966,
          "recall": 0.8493,
          "f1": 0.7654,
          "support": 146,
          "true_positive": 124,
          "false_positive": 54,
          "false_negative": 22
        }
      }
    },
    "isAt": {
      "labels": [
        "TRUE",
        "FALSE"
      ],
      "matrix": [
        [
          21,
          46
        ],
        [
          4,
          167
        ]
      ],
      "rows": "gold",
      "columns": "prediction",
      "total": 238,
      "table": [
        {
          "gold": "TRUE",
          "pred_TRUE": 21,
          "pred_FALSE": 46
        },
        {
          "gold": "FALSE",
          "pred_TRUE": 4,
          "pred_FALSE": 167
        }
      ],
      "accuracy": 0.7899,
      "per_label": {
        "TRUE": {
          "precision": 0.84,
          "recall": 0.3134,
          "f1": 0.4565,
          "support": 67,
          "true_positive": 21,
          "false_positive": 4,
          "false_negative": 46
        },
        "FALSE": {
          "precision": 0.784,
          "recall": 0.9766,
          "f1": 0.8698,
          "support": 171,
          "true_positive": 167,
          "false_positive": 46,
          "false_negative": 4
        }
      }
    }
  }
}