Parameters

Allowed trial hyperparameters (versioned JSON on each trial):

• reg_param — Ridge regularization (default 1.0)
• validation_fraction — chronological holdout (default 0.2)
• standardize_features — train-split mean/std scaling
• feature_exclude — column names to drop (ablation)
• target_column — override training target

Trials are ranked by walk-forward RMSE improvement vs naive plus directional hit rate. Grids expand as a cartesian product, capped at 20 trials per experiment.