Target (y) And Predictor (x) Definition (1.2)

Layer 1 declares which y series is forecast and which source columns are eligible as candidate predictors x. It does not choose y transformations, horizon target construction, x lags, factors, or model-ready feature blocks.

1.2.1 target_structure

Target (y) Definition says whether the recipe has one y series or multiple y series.

Value	Required payload	Meaning
single_target	leaf_config.target	One y series.
multi_target	leaf_config.targets	Multiple y series.

Layer 0 connection:

• one_target_one_method and one_target_compare_methods require target_structure=single_target.

• multiple_targets_one_method and multiple_targets_compare_methods require target_structure=multi_target.

1.2.2 variable_universe

Predictor (x) Universe decides which source columns are eligible as x before Layer 2 representation construction.

Value	Required payload	Meaning
all_variables	none	Use all eligible non-date, non-target source columns.
core_variables	package metadata	Use the package’s core macro subset.
category_variables	category mapping payload	Use columns mapped to a category.
target_specific_variables	leaf_config.target_specific_columns	Use different x sets by target y.
explicit_variable_list	leaf_config.variable_universe_columns	Use an explicit x column list.

Custom data note:

• Custom-only files have no automatic FRED category metadata.

• Use all_variables or explicit_variable_list unless the recipe provides category or target-specific mappings.

Recipe usage:

path:
  1_data_task:
    fixed_axes:
      target_structure: single_target
      variable_universe: explicit_variable_list
    leaf_config:
      target: INDPRO
      horizons: [1, 3, 6]
      variable_universe_columns: [RPI, UNRATE, CPIAUCSL]

Multi-target usage:

path:
  1_data_task:
    fixed_axes:
      target_structure: multi_target
      variable_universe: target_specific_variables
    leaf_config:
      targets: [INDPRO, UNRATE]
      horizons: [1, 3]
      target_specific_columns:
        INDPRO: [RPI, UNRATE, CPIAUCSL]
        UNRATE: [PAYEMS, CLAIMSx, INDPRO]

Boundary

These are not Layer 1 target/predictor-definition choices:

• horizon_target_construction: Layer 2 target representation.

• target_transform and target_normalization: Layer 2 target preprocessing.

• target_lag_block, x_lag_feature_block, factor_feature_block, and feature-block combinations: Layer 2 representation construction.

• forecast_type and forecast_object: Layer 3 forecast-generation contract.

• model_family and benchmark_family: Layer 3 training choices.