comparison_stages
Back to L3.5 | Browse all axes | Browse all options
Axis
comparison_stageson sub-layerL3_5_A_comparison_axis(layerl3_5).
Sub-layer
L3_5_A_comparison_axis
Axis metadata
Default:
'cleaned_vs_features'Sweepable: False
Status: operational
Operational status summary
Operational: 3 option(s)
Future: 0 option(s)
Options
cleaned_vs_features – operational
Compare cleaned panel vs feature-engineered panel (skip raw).
L3.5.A comparison stages cleaned_vs_features.
This option configures the comparison_stages axis on the L3_5_A_comparison_axis sub-layer of L3.5; output is emitted under manifest.diagnostics/l3_5/L3_5_A_comparison_axis/ alongside the other selected views.
When to use
Isolating the L3 contribution when L2’s cleaning is well-trusted.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Related options: raw_vs_cleaned_vs_features, features_only
Last reviewed 2026-05-05 by macroforecast author.
features_only – operational
Inspect feature panel in isolation.
L3.5.A comparison stages features_only.
This option configures the comparison_stages axis on the L3_5_A_comparison_axis sub-layer of L3.5; output is emitted under manifest.diagnostics/l3_5/L3_5_A_comparison_axis/ alongside the other selected views.
When to use
When upstream stages are well-trusted and the focus is on the L3 output’s properties.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Related options: raw_vs_cleaned_vs_features, cleaned_vs_features
Last reviewed 2026-05-05 by macroforecast author.
raw_vs_cleaned_vs_features – operational
Compare raw / cleaned / featurised panels in a 3-way view.
L3.5.A comparison stages raw_vs_cleaned_vs_features.
This option configures the comparison_stages axis on the L3_5_A_comparison_axis sub-layer of L3.5; output is emitted under manifest.diagnostics/l3_5/L3_5_A_comparison_axis/ alongside the other selected views.
When to use
Default broad audit; tracking the panel’s evolution from raw FRED data through to the L3 feature matrix.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Related options: cleaned_vs_features, features_only
Last reviewed 2026-05-05 by macroforecast author.