Known limits and troubleshooting#

Train/test columns differ

Fit once on training data and transform every later split with the same fitted encoder.

Unexpected all-zero category

Check unknown-category policy and whether the value was present at fit time.

Duplicate/collapsed categories

The current maintenance snapshot records an open cross-feature category collapse defect in DummyCodeEncoder. Prefer focused regression coverage for multi-column/category-overlap cases.

Infrequent grouping differs from expectation

An open finding records that the configured grouping path is not fully applied. Treat it as experimental behavior until repaired.