{
  "context": "Evaluates whether the solution applies scikit-learn's probabilistic tooling to deliver uncertainty-aware regression, mixture-based density scoring, and calibrated classification outputs as required by the spec. Checks hinge on correct estimator choices, parameter wiring, and probability handling, not general code style.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "GP regressor",
      "description": "Uses sklearn.gaussian_process.GaussianProcessRegressor with an RBF-like kernel (optionally combined with WhiteKernel) and calls predict(..., return_std=True) to emit paired mean and standard deviation arrays for regression outputs.",
      "max_score": 25
    },
    {
      "name": "Smoothness control",
      "description": "Maps the spec's smoothness argument into the Gaussian process kernel length scale (e.g., kernel.set_params or kernel construction) and reuses it when refitting so lower smoothness values yield more locally varying predictions.",
      "max_score": 15
    },
    {
      "name": "Mixture density",
      "description": "Fits a scikit-learn mixture density estimator such as GaussianMixture or BayesianGaussianMixture with the requested component count and uses score_samples (or equivalent log-density call) to produce per-sample log-likelihoods for anomaly scoring.",
      "max_score": 20
    },
    {
      "name": "Calibrated classifier",
      "description": "Wraps a base classifier that exposes predict_proba in CalibratedClassifierCV (using the spec's method and cv arguments) to train calibrated probability estimates instead of manually postprocessing scores.",
      "max_score": 25
    },
    {
      "name": "Probability outputs",
      "description": "Derives calibrated class probabilities via predict_proba from the calibrated estimator, ensures they are normalized per sample, and maps labels from the argmax of those calibrated probabilities rather than raw scores.",
      "max_score": 15
    }
  ]
}