distribution_metric
Back to L2.5 | Browse all axes | Browse all options
Axis
distribution_metricon sub-layerL2_5_B_distribution_shift(layerl2_5).
Sub-layer
L2_5_B_distribution_shift
Axis metadata
Default:
['mean_change', 'sd_change', 'ks_statistic']Sweepable: False
Status: operational
Operational status summary
Operational: 5 option(s)
Future: 0 option(s)
Options
ks_statistic – operational
Kolmogorov-Smirnov statistic between raw and cleaned distributions.
L2.5.B distribution metric ks_statistic (multi-select axis).
This option configures the distribution_metric axis on the L2_5_B_distribution_shift sub-layer of L2.5; output is emitted under manifest.diagnostics/l2_5/L2_5_B_distribution_shift/ alongside the other selected views.
When to use
Detecting that cleaning has moved the distribution non-trivially; KS > 0.1 is a typical ‘investigate’ threshold.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Related options: mean_change, sd_change, skew_change, kurtosis_change
Last reviewed 2026-05-05 by macroforecast author.
kurtosis_change – operational
Δ-excess-kurtosis before vs after cleaning.
L2.5.B distribution metric kurtosis_change (multi-select axis).
This option configures the distribution_metric axis on the L2_5_B_distribution_shift sub-layer of L2.5; output is emitted under manifest.diagnostics/l2_5/L2_5_B_distribution_shift/ alongside the other selected views.
When to use
Detecting heavy-tail removal; large negative changes confirm successful tail trimming.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Related options: ks_statistic, mean_change, sd_change, skew_change
Last reviewed 2026-05-05 by macroforecast author.
mean_change – operational
Δ-mean before vs after cleaning.
L2.5.B distribution metric mean_change (multi-select axis).
This option configures the distribution_metric axis on the L2_5_B_distribution_shift sub-layer of L2.5; output is emitted under manifest.diagnostics/l2_5/L2_5_B_distribution_shift/ alongside the other selected views.
When to use
Spotting bias introduced by the imputation method (EM-factor can introduce systematic shifts).
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Related options: ks_statistic, sd_change, skew_change, kurtosis_change
Last reviewed 2026-05-05 by macroforecast author.
sd_change – operational
Δ-standard-deviation before vs after cleaning.
L2.5.B distribution metric sd_change (multi-select axis).
This option configures the distribution_metric axis on the L2_5_B_distribution_shift sub-layer of L2.5; output is emitted under manifest.diagnostics/l2_5/L2_5_B_distribution_shift/ alongside the other selected views.
When to use
Detecting variance compression from winsorisation / outlier-replacement.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Related options: ks_statistic, mean_change, skew_change, kurtosis_change
Last reviewed 2026-05-05 by macroforecast author.
skew_change – operational
Δ-skewness before vs after cleaning.
L2.5.B distribution metric skew_change (multi-select axis).
This option configures the distribution_metric axis on the L2_5_B_distribution_shift sub-layer of L2.5; output is emitted under manifest.diagnostics/l2_5/L2_5_B_distribution_shift/ alongside the other selected views.
When to use
Quantifying tail-trimming asymmetry; large skew changes indicate one-sided outlier treatment.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Related options: ks_statistic, mean_change, sd_change, kurtosis_change
Last reviewed 2026-05-05 by macroforecast author.