{
  "context": "Evaluates whether the solution builds the requested mixed-type preprocessing helper using scikit-learn transformers. Scoring checks correct use of built-in imputers, scalers, polynomial feature generators, encoders, and compositional tools to maintain ordering and stability between fit and transform.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "Numeric imputer",
      "description": "Applies sklearn.impute.SimpleImputer (or equivalent) with strategy='median' to numeric columns as part of the fitted pipeline, so transform-time uses the learned medians without manual fill logic.",
      "max_score": 15
    },
    {
      "name": "Numeric scaling",
      "description": "Uses sklearn.preprocessing.StandardScaler on the numeric block after imputation to achieve zero-mean/unit-variance scaling, relying on the scaler's fitted mean_/scale_ during transform.",
      "max_score": 15
    },
    {
      "name": "Polynomial terms",
      "description": "Generates degree-2 polynomial features (squares plus interaction) from the scaled numeric outputs via sklearn.preprocessing.PolynomialFeatures with degree=2 and include_bias=False, and keeps them appended in the output.",
      "max_score": 20
    },
    {
      "name": "Categorical imputer",
      "description": "Uses sklearn.impute.SimpleImputer with strategy='most_frequent' (or equivalent) on categorical columns before encoding so missing categories are filled consistently across fit/transform.",
      "max_score": 10
    },
    {
      "name": "One-hot encoding",
      "description": "Applies sklearn.preprocessing.OneHotEncoder with handle_unknown='ignore' and dense output (sparse=False or toarray()) to the categorical block so unseen categories at transform time do not raise and produce all-zero slices.",
      "max_score": 20
    },
    {
      "name": "Column composition",
      "description": "Combines numeric and categorical pipelines with sklearn.compose.ColumnTransformer (optionally wrapped in sklearn.pipeline.Pipeline), preserving output order (scaled numerics, polynomial terms, then encoded categoricals) and exposing deterministic feature names via get_feature_names_out.",
      "max_score": 20
    }
  ]
}