Localization Headers

Create language-specific views of a dataset where only the headers (and optionally the data dictionary text) are translated. The row data is unchanged unless the user asks for value-level localisation separately.

When to invoke

• User wants to share a dataset with non-English-speaking analysts and needs translated column headers.
• User wants a multilingual data dictionary derived from one canonical version.
• Preparing a dataset for a multi-region team where each locale reads its own header language.
• User says "localise this dataset" or "make a Hebrew/Arabic/… version".

Design principles

• Canonical dataset stays untouched. Localisation produces sibling files; it does not mutate the source.
• Column order and content are preserved across all language variants so joins and diffs work by position.
• One source of truth for the mapping. A single header_translations.{csv,yaml} file maps canonical headers to each locale. All language variants and the localised dictionaries are generated from this file. Re-running should be idempotent.
• Low friction. The user provides target language(s); everything else is automated. The user only has to approve the proposed translations.

Procedure