{
  "context": "Evaluates whether the solution leverages scikit-learn's model selection stack to tune and assess a classifier using cross-validation, hyperparameter search, metrics, and learning-curve diagnostics. Checks focus on the proper use of built-in splitters, search classes, scoring configuration, and learning curve utilities to produce the required summary outputs.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "Splitter usage",
      "description": "Uses a label-aware scikit-learn cross-validation splitter (e.g., StratifiedKFold or GroupKFold when groups provided) with the requested fold count and random_state.",
      "max_score": 20
    },
    {
      "name": "CV scoring",
      "description": "Computes per-fold and mean scores with the provided scoring string via scikit-learn utilities like cross_val_score or the cv results from a search object, matching the fold configuration.",
      "max_score": 20
    },
    {
      "name": "Search setup",
      "description": "Runs a scikit-learn hyperparameter search (GridSearchCV, RandomizedSearchCV, or HalvingGridSearchCV) over the supplied grid using the chosen splitter, scoring, and random_state, and refits the best estimator.",
      "max_score": 25
    },
    {
      "name": "Learning curve",
      "description": "Generates training sizes and paired train/validation scores using sklearn.model_selection.learning_curve with the same splitter/scoring and honors provided learning_curve_sizes or defaults to at least five evenly spaced fractions.",
      "max_score": 20
    },
    {
      "name": "Result bundle",
      "description": "Returns best_params, best_score, fold_scores, cv_mean_score, learning_curve arrays, and the refit estimator consistent with search and learning_curve outputs.",
      "max_score": 15
    }
  ]
}