{
  "best-base_weighted_ce-ep5_vs_GPT-5.4": {
    "cat_macro_f1": 0.9360575579144376,
    "cat_weighted_f1": 0.9360564701876355,
    "cat_macro_precision": 0.9336791798534633,
    "cat_macro_recall": 0.9414080218768329,
    "cat_mcc": 0.9248088496355107,
    "cat_auc": 0.991343460842945,
    "cat_ece": 0.0441274690628052,
    "cat_confusion_matrix": [
      [
        220,
        0,
        8,
        0,
        1,
        0,
        1
      ],
      [
        0,
        86,
        0,
        0,
        1,
        1,
        0
      ],
      [
        1,
        0,
        143,
        1,
        5,
        0,
        0
      ],
      [
        0,
        0,
        2,
        133,
        0,
        1,
        0
      ],
      [
        6,
        0,
        7,
        18,
        165,
        1,
        1
      ],
      [
        0,
        3,
        1,
        8,
        2,
        207,
        0
      ],
      [
        0,
        0,
        0,
        1,
        6,
        1,
        169
      ]
    ],
    "cat_f1_BoardGov": 0.962800875273523,
    "cat_prec_BoardGov": 0.9691629955947136,
    "cat_recall_BoardGov": 0.9565217391304348,
    "cat_f1_Incident": 0.9717514124293786,
    "cat_prec_Incident": 0.9662921348314607,
    "cat_recall_Incident": 0.9772727272727273,
    "cat_f1_Manageme": 0.9196141479099679,
    "cat_prec_Manageme": 0.8881987577639752,
    "cat_recall_Manageme": 0.9533333333333334,
    "cat_f1_NoneOthe": 0.8956228956228957,
    "cat_prec_NoneOthe": 0.8260869565217391,
    "cat_recall_NoneOthe": 0.9779411764705882,
    "cat_f1_RiskMana": 0.873015873015873,
    "cat_prec_RiskMana": 0.9166666666666666,
    "cat_recall_RiskMana": 0.8333333333333334,
    "cat_f1_Strategy": 0.9583333333333334,
    "cat_prec_Strategy": 0.981042654028436,
    "cat_recall_Strategy": 0.9366515837104072,
    "cat_f1_Third-Pa": 0.9712643678160919,
    "cat_prec_Third-Pa": 0.9883040935672515,
    "cat_recall_Third-Pa": 0.9548022598870056,
    "cat_kripp_alpha": 0.9243601922903683,
    "spec_macro_f1": 0.5970357338282843,
    "spec_weighted_f1": 0.7040798408451929,
    "spec_macro_precision": 0.7225196233593912,
    "spec_macro_recall": 0.6139005306639329,
    "spec_mcc": 0.6138700055328291,
    "spec_auc": 0.9498756282617218,
    "spec_ece": 0.1652249880135059,
    "spec_confusion_matrix": [
      [
        596,
        6,
        11,
        5
      ],
      [
        105,
        46,
        9,
        8
      ],
      [
        14,
        6,
        52,
        135
      ],
      [
        4,
        0,
        3,
        200
      ]
    ],
    "spec_f1_L1Generi": 0.8915482423335827,
    "spec_prec_L1Generi": 0.8289290681502086,
    "spec_recall_L1Generi": 0.9644012944983819,
    "spec_f1_L2Domain": 0.40707964601769914,
    "spec_prec_L2Domain": 0.7931034482758621,
    "spec_recall_L2Domain": 0.27380952380952384,
    "spec_f1_L3Firm-S": 0.36879432624113473,
    "spec_prec_L3Firm-S": 0.6933333333333334,
    "spec_recall_L3Firm-S": 0.25120772946859904,
    "spec_f1_L4Quanti": 0.7207207207207207,
    "spec_prec_L4Quanti": 0.5747126436781609,
    "spec_recall_L4Quanti": 0.966183574879227,
    "spec_qwk": 0.8757404773441285,
    "spec_mae": 0.2975,
    "spec_kripp_alpha": 0.8479072400833478,
    "total_time_s": 6.695346015971154,
    "num_samples": 1200,
    "avg_ms_per_sample": 5.579455013309295,
    "combined_macro_f1": 0.766546645871361
  },
  "best-base_weighted_ce-ep5_vs_Opus-4.6": {
    "cat_macro_f1": 0.9280167387549427,
    "cat_weighted_f1": 0.9273898648954128,
    "cat_macro_precision": 0.9223465490796974,
    "cat_macro_recall": 0.9382296607170699,
    "cat_mcc": 0.9162751746063641,
    "cat_auc": 0.992382433433919,
    "cat_ece": 0.04689237485329312,
    "cat_confusion_matrix": [
      [
        209,
        0,
        2,
        1,
        1,
        0,
        1
      ],
      [
        0,
        78,
        0,
        0,
        1,
        0,
        0
      ],
      [
        4,
        0,
        147,
        2,
        4,
        0,
        1
      ],
      [
        0,
        0,
        1,
        139,
        1,
        0,
        0
      ],
      [
        13,
        1,
        10,
        15,
        168,
        1,
        5
      ],
      [
        1,
        10,
        1,
        4,
        3,
        209,
        0
      ],
      [
        0,
        0,
        0,
        0,
        2,
        1,
        164
      ]
    ],
    "cat_f1_BoardGov": 0.9478458049886621,
    "cat_prec_BoardGov": 0.920704845814978,
    "cat_recall_BoardGov": 0.9766355140186916,
    "cat_f1_Incident": 0.9285714285714286,
    "cat_prec_Incident": 0.8764044943820225,
    "cat_recall_Incident": 0.9873417721518988,
    "cat_f1_Manageme": 0.9216300940438872,
    "cat_prec_Manageme": 0.9130434782608695,
    "cat_recall_Manageme": 0.930379746835443,
    "cat_f1_NoneOthe": 0.9205298013245033,
    "cat_prec_NoneOthe": 0.8633540372670807,
    "cat_recall_NoneOthe": 0.9858156028368794,
    "cat_f1_RiskMana": 0.8549618320610687,
    "cat_prec_RiskMana": 0.9333333333333333,
    "cat_recall_RiskMana": 0.7887323943661971,
    "cat_f1_Strategy": 0.9521640091116174,
    "cat_prec_Strategy": 0.990521327014218,
    "cat_recall_Strategy": 0.9166666666666666,
    "cat_f1_Third-Pa": 0.9704142011834319,
    "cat_prec_Third-Pa": 0.9590643274853801,
    "cat_recall_Third-Pa": 0.9820359281437125,
    "cat_kripp_alpha": 0.9154955768233572,
    "spec_macro_f1": 0.5957642708821952,
    "spec_weighted_f1": 0.693005282664721,
    "spec_macro_precision": 0.731933400476396,
    "spec_macro_recall": 0.6249872364065566,
    "spec_mcc": 0.6143201053040909,
    "spec_auc": 0.9470735892830423,
    "spec_ece": 0.18189165468017254,
    "spec_confusion_matrix": [
      [
        592,
        4,
        4,
        5
      ],
      [
        92,
        42,
        6,
        5
      ],
      [
        35,
        12,
        63,
        150
      ],
      [
        0,
        0,
        2,
        188
      ]
    ],
    "spec_f1_L1Generi": 0.8942598187311178,
    "spec_prec_L1Generi": 0.8233657858136301,
    "spec_recall_L1Generi": 0.9785123966942149,
    "spec_f1_L2Domain": 0.41379310344827586,
    "spec_prec_L2Domain": 0.7241379310344828,
    "spec_recall_L2Domain": 0.2896551724137931,
    "spec_f1_L3Firm-S": 0.3761194029850746,
    "spec_prec_L3Firm-S": 0.84,
    "spec_recall_L3Firm-S": 0.2423076923076923,
    "spec_f1_L4Quanti": 0.6988847583643123,
    "spec_prec_L4Quanti": 0.5402298850574713,
    "spec_recall_L4Quanti": 0.9894736842105263,
    "spec_qwk": 0.872110225054491,
    "spec_mae": 0.3075,
    "spec_kripp_alpha": 0.8502616991488389,
    "total_time_s": 6.695346015971154,
    "num_samples": 1200,
    "avg_ms_per_sample": 5.579455013309295,
    "combined_macro_f1": 0.7618905048185689
  }
}