Recipe Layer Contract

A recipe is a top-level YAML mapping. Main layers are L0-L8; diagnostic layers are L1.5-L4.5. L3, L4, and L7 use DAG nodes plus sinks. L1, L2, L5, L6, and L8 primarily use fixed_axes plus optional leaf_config.

This complete recipe is runnable on a stock install:

0_meta:
  fixed_axes: {failure_policy: fail_fast, reproducibility_mode: seeded_reproducible}
  leaf_config: {random_seed: 42}

1_data:
  fixed_axes: {custom_source_policy: custom_panel_only, frequency: monthly, horizon_set: custom_list}
  leaf_config:
    target: y
    target_horizons: [1]
    custom_panel_inline:
      date: [2018-01-01, 2018-02-01, 2018-03-01, 2018-04-01, 2018-05-01, 2018-06-01, 2018-07-01, 2018-08-01, 2018-09-01, 2018-10-01, 2018-11-01, 2018-12-01]
      y:  [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0]
      x1: [0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0, 4.5, 5.0, 5.5, 6.0]

2_preprocessing:
  fixed_axes: {transform_policy: no_transform, outlier_policy: none, imputation_policy: none_propagate, frame_edge_policy: keep_unbalanced}

3_feature_engineering:
  nodes:
    - {id: src_X, type: source, selector: {layer_ref: l2, sink_name: l2_clean_panel_v1, subset: {role: predictors}}}
    - {id: src_y, type: source, selector: {layer_ref: l2, sink_name: l2_clean_panel_v1, subset: {role: target}}}
    - {id: lag_x, type: step, op: lag, params: {n_lag: 1}, inputs: [src_X]}
    - {id: y_h, type: step, op: target_construction, params: {mode: point_forecast, method: direct, horizon: 1}, inputs: [src_y]}
  sinks:
    l3_features_v1: {X_final: lag_x, y_final: y_h}
    l3_metadata_v1: auto

4_forecasting_model:
  nodes:
    - id: src_X
      type: source
      selector: {layer_ref: l3, sink_name: l3_features_v1, subset: {component: X_final}}
    - id: src_y
      type: source
      selector: {layer_ref: l3, sink_name: l3_features_v1, subset: {component: y_final}}
    - id: fit_ridge
      type: step
      op: fit_model
      params:
        family: ridge
        alpha: 1.0
        forecast_strategy: direct
        training_start_rule: expanding
        refit_policy: every_origin
        search_algorithm: none
        min_train_size: 6
      is_benchmark: true
      inputs: [src_X, src_y]
    - id: predict_ridge
      type: step
      op: predict
      inputs: [fit_ridge, src_X]
  sinks:
    l4_forecasts_v1: predict_ridge
    l4_model_artifacts_v1: fit_ridge
    l4_training_metadata_v1: auto