Source, Frequency, And Forecast-Time Information (1.1)

Layer 1 starts with the source frame. The first choice is Data Source Mode: use FRED data only, custom data only, or FRED data plus custom columns. After that, close the analysis frequency and forecast-time information contract.

Section	Axis / payload	User-facing name	Role
1.1.1	`custom_source_policy`	Data Source Mode	FRED data only, custom data only, or FRED plus custom data
1.1.2	`dataset`	FRED Source Panel	Which FRED panel to load; hidden for custom-only studies
1.1.3	`custom_source_path`	Custom file path	Study-specific file path in `leaf_config`; not a Navigator axis
1.1.4	`frequency`	Analysis Frequency	Monthly or quarterly Layer 1 frame
1.1.5	`information_set_type`	Data Revision / Vintage Regime	Final revised data or pseudo-OOS on revised data

At a glance.

custom_source_policy = official_only is the default.
dataset is required only when the source mode uses FRED data.
custom_panel_only should not expose a FRED Source Panel choice; it needs frequency and leaf_config.custom_source_path.
official_plus_custom loads a FRED Source Panel and appends custom columns that already match the selected frequency.
information_set_type controls revision/vintage status, not publication lag. Publication lag is Data Handling Policies.

data_domain was removed because every built-in FRED source already implies a macro domain through source metadata.

1.1.1 `custom_source_policy`

Data Source Mode decides what defines the Layer 1 panel.

Value	Label	Meaning
`official_only`	FRED Data Only (Default)	Load one FRED Source Panel.
`custom_panel_only`	Custom Data Only	Load a custom file as the source panel.
`official_plus_custom`	FRED Data + Custom Data	Load a FRED Source Panel and append custom columns.

Custom modes require:

path:
  1_data_task:
    leaf_config:
      custom_source_path: /path/to/panel.csv

1.1.2 `dataset`

FRED Source Panel is active only when custom_source_policy is official_only or official_plus_custom.

Value	Loader	Frequency rule
`fred_md`	`macroforecast.raw.load_fred_md`	monthly
`fred_qd`	`macroforecast.raw.load_fred_qd`	quarterly
`fred_sd`	`macroforecast.raw.load_fred_sd`	user chooses monthly or quarterly
`fred_md+fred_sd`	composite loader	monthly
`fred_qd+fred_sd`	composite loader	quarterly

Do not use dataset as a fake schema selector for custom-only data. Custom files are accepted by file shape: parseable dates, numeric series columns, and rows at the selected frequency.

1.1.3 Custom File Path Contract

custom_source_path is a leaf_config payload because it is study-specific, not a reusable enum axis.

Parser inference:

.csv -> macroforecast.raw.load_custom_csv
.parquet or .pq -> macroforecast.raw.load_custom_parquet

File shape:

first CSV column, first Parquet column, or Parquet index must be parseable as dates;
remaining columns must be numeric series;
monthly custom data must have monthly rows;
quarterly custom data must have quarterly rows;
for custom-only data, the target y column must exist in the file;
candidate predictor x columns are non-target numeric columns unless variable_universe narrows them.

The package can validate file parseability and enum choices. It cannot certify economic meaning, release timing, vintage discipline, or whether a T-code row was stripped or applied correctly. Treat the file path and file hash as provenance.

1.1.4 `frequency`

Analysis Frequency is the final calendar frequency of the Layer 1 source frame.

Value	Meaning
`monthly`	Monthly source frame. Required for FRED-MD and MD+SD; allowed for FRED-SD and custom-only.
`quarterly`	Quarterly source frame. Required for FRED-QD and QD+SD; allowed for FRED-SD and custom-only.

Runtime conversion is available for FRED-SD monthly/quarterly conversion and is recorded in provenance. Custom files must already be at the selected frequency.

1.1.5 `information_set_type`

Data Revision / Vintage Regime decides which version of the data is allowed at each forecast origin.

Value	Label	Meaning
`final_revised_data`	Final Revised Data (Default)	Use the latest revised values.
`pseudo_oos_on_revised_data`	Pseudo-OOS on Revised Data	Use revised data with pseudo out-of-sample masking.

This is not the publication-lag rule. Use release_lag_rule when the question is “when was this series published and usable?”

Recipe Examples

FRED-MD:

path:
  1_data_task:
    fixed_axes:
      custom_source_policy: official_only
      dataset: fred_md
      information_set_type: final_revised_data
    leaf_config:
      target: INDPRO
      horizons: [1, 3, 6]

Custom-only monthly CSV:

path:
  1_data_task:
    fixed_axes:
      custom_source_policy: custom_panel_only
      frequency: monthly
      information_set_type: final_revised_data
    leaf_config:
      target: INDPRO
      horizons: [1, 3]
      custom_source_path: /path/to/custom_monthly_panel.csv

FRED-QD plus custom quarterly columns:

path:
  1_data_task:
    fixed_axes:
      custom_source_policy: official_plus_custom
      dataset: fred_qd
      frequency: quarterly
      information_set_type: final_revised_data
    leaf_config:
      target: GDPC1
      horizons: [1, 2, 4]
      custom_source_path: /path/to/custom_quarterly_columns.parquet

Takeaways

Choose Data Source Mode first.
Choose FRED Source Panel only when the source mode uses FRED data.
Choose Analysis Frequency explicitly for standalone FRED-SD and custom-only data.
Treat Data Revision / Vintage Regime and Publication Lag Rule as different decisions.

Source, Frequency, And Forecast-Time Information (1.1)

1.1.1 custom_source_policy

1.1.2 dataset