Categorical cleanup
Standardize categories with rules you can read and audit
Labels such as United States, USA, U.S., and usa may represent one category, but a safe cleanup should make the mapping visible. Controlled case normalization and explicit value replacement keep that decision reviewable.
Start with deterministic text normalization
Whitespace trimming and case normalization resolve many superficial variants without changing the semantic label. The plan targets named columns and specifies the requested case behavior, so the transformation remains clear before execution.
For known semantic aliases, replace-values uses an explicit from-to mapping and a case-sensitivity option. Free-form code and remote lookup URLs are rejected by the rule schema.
Keep ambiguous categories visible
Not every similar label is equivalent. A system should not assume that two product codes, regions, diagnoses, or account statuses belong together without domain context. The current real workflow proposes controlled mappings but waits for user review.
The planned data dictionary and domain-notes editor can make those decisions easier to explain. It remains development-preview data in this phase and is not persisted or sent to the planning model.
- Trim leading and trailing whitespace
- Normalize lower, upper, or title case
- Replace only explicitly mapped values
- Review target columns and impact before running
Handle new categories as schema drift, not a silent fix
When future recipes are applied to new files, unseen categories must be surfaced during drift review. A new label may be accepted, mapped, or used to create a new recipe version, but it should not be silently folded into an existing category.
Recipe persistence and cross-file mapping are not connected yet. The public product accurately labels that workflow as coming soon while the real run report continues to show only transformations and provenance backed by stored execution data.