models/vector_calibration_evaluation.json

{
  "baseline": "shipped global temperature calibration",
  "method": "regularized five-parameter multiclass vector scaling",
  "rolling_origin_regularization_scores": {
    "0.0001": 0.7525188824528779,
    "0.001": 0.7526886773719569,
    "0.01": 0.7541321350490496,
    "0.1": 0.7578710034159443
  },
  "selected_regularization": 0.0001,
  "parameters": [
    1.0484497532436863,
    0.8770152538723214,
    0.9388567787680596,
    0.8290372459085292,
    0.49591824484502284
  ],
  "slices": {
    "selection_descriptive": {
      "baseline": {
        "n": 7914,
        "log_loss": 0.7917934118271145,
        "brier": 0.4641057448243238,
        "rps": 0.1471617663399534,
        "ece": 0.00846231279824596
      },
      "candidate": {
        "n": 7914,
        "log_loss": 0.7651349976317772,
        "brier": 0.44767783933088434,
        "rps": 0.14003929365045623,
        "ece": 0.009059596794553856
      },
      "blocked_log_loss_difference_95": [
        -0.032996290530500585,
        -0.020046464365746766
      ]
    },
    "robustness_2026_viewed": {
      "baseline": {
        "n": 351,
        "log_loss": 0.7931215868621926,
        "brier": 0.46119749518248465,
        "rps": 0.1453477516943903,
        "ece": 0.04067188143421708
      },
      "candidate": {
        "n": 351,
        "log_loss": 0.7638626804396227,
        "brier": 0.44661751123246113,
        "rps": 0.13839649327242182,
        "ece": 0.035138558590492315
      },
      "blocked_log_loss_difference_95": [
        -0.05139728352686812,
        -0.005637635590483102
      ]
    },
    "robustness_2026_neutral_viewed": {
      "baseline": {
        "n": 143,
        "log_loss": 0.7916431257734952,
        "brier": 0.45920689749839105,
        "rps": 0.1474630439153726,
        "ece": 0.10246247217248158
      },
      "candidate": {
        "n": 143,
        "log_loss": 0.6571008523753717,
        "brier": 0.37555349693483997,
        "rps": 0.10922584831530416,
        "ece": 0.16122760848044432
      },
      "blocked_log_loss_difference_95": [
        -0.14603544930037682,
        -0.12198283454688964
      ]
    }
  },
  "research_gate_passed": false,
  "production_adopted": false,
  "claim_boundary": "The 2026 slice has already been viewed and used in prior model decisions. It is a robustness diagnostic, not pristine confirmation. Production adoption requires a future pre-registered evaluation period."
}