Data Policies#

Data-policy axes live in Layer 1 and Layer 2.

Layer 1 policies decide what the recipe means before preprocessing:

Axis	Valid values
`missing_availability`	`require_complete_rows`, `keep_available_rows`, `impute_predictors_only`, `zero_fill_leading_predictor_gaps`
`raw_missing_policy`	`preserve_raw_missing`, `zero_fill_leading_predictor_missing_before_tcode`, `impute_raw_predictors`, `drop_raw_missing_rows`
`raw_outlier_policy`	`preserve_raw_outliers`, `winsorize_raw`, `iqr_clip_raw`, `mad_clip_raw`, `zscore_clip_raw`, `set_raw_outliers_to_missing`
`release_lag_rule`	`ignore_release_lag`, `fixed_lag_all_series`, `series_specific_lag`
`contemporaneous_x_rule`	`allow_same_period_predictors`, `forbid_same_period_predictors`
`official_transform_policy`	`apply_official_tcode`, `keep_official_raw_scale`
`official_transform_scope`	`target_only`, `predictors_only`, `target_and_predictors`, `none`

The current default for same-period predictors is allow_same_period_predictors.

Layer 2 applies cleaning after L1: transform, outlier handling, imputation, and frame-edge trimming.