Legacy Callable Coverage#

Back to reference

This page tracks the old runtime/reference surface against the current callable-first package. It is a migration audit, not a public recipe grammar.

Current verdict: the old statistical functionality is represented as callable modules. The intentionally removed surface is YAML/runtime/registry plumbing that will be rebuilt later as a wrapper around the callable API.

Last updated: 2026-06-01.

Intentional Removals#

These old surfaces are not gaps.

Old surface	Current decision	Reason
Old recipe schemas, validators, and parse helpers	Removed for now	Recipe/YAML support will be rebuilt as a thin wrapper over callables.
Old runtime dispatch nodes and materializers	Removed for now	Runner and direct callables now own execution.
Old compatibility shims and registries	Removed	Restoring compatibility shims would preserve the structure we are replacing.
`api/functions/`, `functions/` compatibility namespaces	Removed	Public API is now module-based: `data`, `preprocessing`, `feature_engineering`, etc.
`option_docs.py` generated option tables	Removed	Function-level docs now document input, output, defaults, and choices directly.
Recipe/gallery contract pages	Removed	Recipe docs will be rewritten after the callable API stabilizes.
Paper-method recipe constructors	Removed for now	Paper methods should be expressed as callable combinations or future recipe presets.
Holiday feature op	Intentionally omitted	Package targets monthly and quarterly macro panels; public-holiday calendars are not a core axis.

Covered Callable Surface#

Area	Current module	Coverage
Global defaults	`macroforecast.meta`	Package-level config, seed/default management, context manager.
Data panels and metadata	`macroforecast.data`	Canonical pandas panel, metadata attachment, custom CSV/parquet loading, FRED-MD/QD/SD loaders, vintage listing, frequency inference/alignment, Chow-Lin disaggregation, same-period policy, regime metadata.
Data summary and comparison	`macroforecast.data_analysis`	Single-panel summary and before/after analysis.
Preprocessing	`macroforecast.preprocessing`	FRED-style transform codes, FRED-SD code suggestions, outlier handling, imputation, standardization, frame-edge handling, frequency alignment, reusable fit/transform specs.
Filters and smoothers	`macroforecast.filters`	Direct one-series HP, Hamilton, Savitzky-Golay, wavelet-style, and AlbaMA filters/smoothers with metadata.
Feature construction	`macroforecast.feature_engineering`	Direct, average, and path targets; lags; rolling means; filter-to-feature wrappers including AlbaMA adaptive moving-average features with R-code-aligned terminal-node weights; MARX/MAF; mixed-frequency lags; time/season dummies; transforms; polynomial/interactions; PCA, group PCA, sparse PCA, varimax, PLS, SIR, DFM; random projection/Nystroem; selection methods; reusable feature specs.
Feature analysis	`macroforecast.feature_analysis`	Overview, correlation, target correlation, factor diagnostics, lag diagnostics, MARX diagnostics, AlbaMA/effective-window weight summaries, stage shift, selection stability/similarity, custom diagnostics.
Window and policies	`macroforecast.window`	Estimation/validation/test windows, expanding/rolling/fixed modes, blocked splits, origin generation, retrain/retune timing policies, stage panels.
Model selection	`macroforecast.model_selection`	Fixed/grid/random/Bayesian/genetic/custom search specs, distributions, parameter selection over supplied windows.
Models	`macroforecast.models`	Linear/penalized/group/adaptive, supervised factor models, PCR-style composites through factor models, SVR, tree, MARS, Macro Random Forest, MIDAS/mixed-frequency, AR/VAR/BVAR/FAR/FAVAR/DFM, volatility, torch-backed NN/LSTM/GRU/Transformer/HNN, model-owned search spaces, save/load helpers.
Model ensemble	`macroforecast.model_ensemble`	Fit-time member-model composition: bagging, subagging, random subspace, stacking, Super Learner, Booging, and model-ensemble-owned search spaces.
Forecast runner	`macroforecast.forecasting`	Runner composition across windows, preprocessing specs, feature specs, model specs, model ensembles, model selection, direct/recursive/path-average policies, forecast-output combinations.
Metrics	`macroforecast.metrics`	Point, relative, directional, interval, density, quantile, QLIKE, Theil, OOS R-squared, ranking.
Tests	`macroforecast.tests`	DM/GW/DMP/HN, nested/encompassing tests, CPA, exact MCS, SPA/Reality/StepM benchmark tests, blocked OOB reality check, density/interval tests, PIT diagnostics, residual diagnostics, custom tests.
Evaluation reports	`macroforecast.evaluation`	Multi-slice evaluation, OOS filtering, benchmark comparisons, regime scoring, error decomposition, aggregation and ranking.
Forecast analysis	`macroforecast.forecast_analysis`	Forecast overview, fitted-vs-actual, residual views, rolling loss, scale view, tuning traces, coefficient/parameter stability, DFM diagnostics, ensemble diagnostics, stage update trace.
Interpretation	`macroforecast.interpretation`	Native coefficients/importances, permutation/Strobl/LOFO, PDP/ALE/Friedman-H, SHAP variants, anatomy/PBSV forecast-accuracy attribution, forecast decomposition, cumulative contribution, group/lineage/transformation attribution, OLS attention, MRF GTVP, VAR IRF/FEVD/historical decomposition, neural attribution, LSTM hidden states, custom interpretation.
Output	`macroforecast.output`	Output table generation, selection/naming/bundling, artifact writing, manifest, hashes, provenance, gzip/zip compression.
Reporting	`macroforecast.reporting`	Report-table formatting, accuracy/model-comparison/forecast-test presets, LaTeX/HTML/Markdown rendering, figure-ready data, report bundles.
Custom extensions	Per-module custom hooks	Custom datasets, preprocessing, feature steps, models, model ensembles, stage policies, combinations, diagnostics, interpretation, and artifacts. See `reference/custom/`; there is no separate `macroforecast.custom` module.

Remaining Real Gaps#

These are not old runtime cleanup items. They are possible future implementation work on top of the current callable API.

Priority	Gap	Current status	Proposed handling
Medium	Replication/report package builder	`output` writes artifacts and `reporting` renders tables, but there is no one-call replication package assembler.	Add as future callable after output/reporting settle; likely `reporting` or a small package module, not `output`.
Low	External reference-verification expansions	`tests/reference/` now contains core anchors, but not every paper formula against external source code or known-DGP simulations.	Expand the reference suite incrementally as source-code fixtures or accepted known-DGP designs are added.
Low	Future YAML/recipe wrapper	Intentionally absent.	Rebuild later on top of the callable API, with no old runtime registry dependency.

No Current Action Needed#

These areas looked like gaps in the old tree but are now covered or intentionally excluded:

Item	Status
Legacy diagnostic ops	Covered by `feature_analysis`, `forecast_analysis`, and `data_analysis`.
Old output ops	Covered by `output` plus the new `reporting` module.
Legacy MIDAS feature/model split	Covered by `mixed_frequency_lags()` and MIDAS model callables.
Legacy Chow-Lin op	Covered by `data.chow_lin_disaggregate()` and `data.align_frequency(..., quarterly_to_monthly="chow_lin")`.
Legacy blocked OOB reality check	Covered by `tests.blocked_oob_reality_check()`.
Full iterative MCS	Covered by exact `tests.model_confidence_set()`; `tests.iterative_model_confidence_set()` remains a descriptive alias.
Reporting presets	Covered by `reporting.accuracy_table()`, `reporting.model_comparison_table()`, and `reporting.forecast_test_table()`.
Holiday features	Intentionally excluded for monthly/quarterly macro workflows.