Outputs and diagnostics

What appears after `fit`

After fitting a Binner, the public API exposes friendly artifacts:

binning_table()
feature_binning_table() and get_binning_table()
summary()
score_details()
score_table()
report()
audit_table()
diagnostics()
plot_stability()
plot_bad_rate_over_time()
plot_bad_rate_heatmap()
plot_bin_share_over_time()
plot_score_components()
export_binnings_json()
export_audit_report()
export_bundle()

Useful post-fit attributes are also available, such as:

binning_table_
summary_
score_details_
score_table_
audit_table_
report_
metadata_
score_
comparison_score_

Main tables

binning_table() shows final cuts or bins. score_table() is the short table for explaining the objective, weights, normalized components, and raw components. audit_table() combines cuts, IV, temporal score, penalties, coverage, rare bins, reversals, and summarized rationale.

Temporal diagnostics

Use diagnostics(kind="bin") to open detail by bin and period. Use diagnostics(kind="variable") for the aggregated temporal summary by variable.

This is the best entry point to investigate:

coverage
event-rate volatility
WoE volatility
bin share
ranking reversals
monotonicity breaks

Metadata

metadata_ includes the riskbands version, strategy, score_strategy, normalization_strategy, woe_shrinkage_strength, weights, target_name, time_col, and fitted features.

Auditable export

export_binnings_json(path) generates a single JSON with metadata, score weights, bins by feature, summary, score details, and feature-level audit.

export_audit_report(path) generates audit_report.html: a standalone narrative HTML report with embedded CSS, no external assets, print-friendly styling, and an audit/model-risk audience.

export_bundle(path) generates readable JSON, CSVs, feature-level tables, optional Parquet when an engine is available, and audit_report.html by default. Use export_bundle(path, include_audit_report=False) only when the bundle must omit the narrative HTML.

Bundles also persist the missing-values trail when available: missing_policy, effective_missing_policy, missing_profile, missing_decision_log, missing_merge_criterion, missing_merge_fallback, missing_merge_candidates, and missing_merge_map.

These fields record the missing-values treatment decision. They do not represent opaque imputation.

Use missing_profile_ to review missing volume, share, and event rate by variable. Use missing_decision_log_ to see whether the action was to preserve standard, create a Missing bin with separate_bin, block with forbid, or route missing values with merge. For merge, use missing_merge_candidates_ to review candidates and distances and missing_merge_map_ to see the learned destination.

Dedicated guide: Missing policy.

Narrative audit report

audit_report.html explains configuration, variables, missing policies, merge decisions, evaluated candidates, validation, alerts, bundle inventory, and limitations. For merge, the text makes clear that the decision was learned during fit and reused during transform, without recalculating event rate or WoE on application data.

The HTML is self-contained and can be printed or exported to PDF by the browser. It organizes technical evidence, but it does not replace independent formal validation and does not constitute regulatory certification.

Fast score reading

standard: higher raw score is better
stable: lower raw score is better

For a consolidated scale across strategies, also look at objective_preference_score.

Example

binner.fit(df, y="target", column="score", time_col="month")

table = binner.binning_table()
score_table = binner.score_table()
audit_table = binner.audit_table()

binner.export_binnings_json("artifacts/riskbands_binnings.json")
binner.export_audit_report("artifacts/audit_report.html")
binner.export_bundle("artifacts/run_2026_04_14")