Problem Statement
Adding a new curated language to SkillBridge currently means touching the dictionary, locale metadata, the language selector, tests, validation scripts, and documentation together. That coupling makes it unclear what "adding a language" actually requires, and raises the risk that a new locale ships below the quality bar of the existing curated dictionaries.
Proposed Solution
Add one new curated language end to end, chosen on evidence rather than preference — current coverage gaps, project conventions, and whether the terminology can realistically be maintained over time.
The work should cover, in one coherent change:
- the dictionary itself
- locale metadata
- language selector integration
- tests
- validation script coverage
- documentation
Quality bar: match the existing curated dictionaries. AI/ML/Claude-specific terminology deserves particular scrutiny, since that is where a generic translation engine most reliably produces wrong output. The new locale must pass the same validation the existing ones do — check-glossary, check-i18n-keys, check-locale-contamination, check-dict-coverage, and validate-translations.
Whichever language is chosen, the rationale for choosing it belongs in the PR description.
Alternatives Considered
Bulk-generating a dictionary from machine translation output. This does not meet the curated bar and should not be the shape of this work — see below.
Additional Context
An exploratory prototype was attempted locally in August (Dutch / nl was the candidate explored). Its working tree was lost before it could be pushed, so this is filed as a proposal rather than a PR.
A lesson from that attempt is worth recording: it drifted toward mass generation — helper scripts named generate-nl-draft.js and fill-nl-machine.js were created to populate the dictionary from machine output. That directly contradicts what "curated" means here. If scripting is used at all, it should scaffold structure and surface gaps for human review, not fill in terminology.
Problem Statement
Adding a new curated language to SkillBridge currently means touching the dictionary, locale metadata, the language selector, tests, validation scripts, and documentation together. That coupling makes it unclear what "adding a language" actually requires, and raises the risk that a new locale ships below the quality bar of the existing curated dictionaries.
Proposed Solution
Add one new curated language end to end, chosen on evidence rather than preference — current coverage gaps, project conventions, and whether the terminology can realistically be maintained over time.
The work should cover, in one coherent change:
Quality bar: match the existing curated dictionaries. AI/ML/Claude-specific terminology deserves particular scrutiny, since that is where a generic translation engine most reliably produces wrong output. The new locale must pass the same validation the existing ones do —
check-glossary,check-i18n-keys,check-locale-contamination,check-dict-coverage, andvalidate-translations.Whichever language is chosen, the rationale for choosing it belongs in the PR description.
Alternatives Considered
Bulk-generating a dictionary from machine translation output. This does not meet the curated bar and should not be the shape of this work — see below.
Additional Context
An exploratory prototype was attempted locally in August (Dutch /
nlwas the candidate explored). Its working tree was lost before it could be pushed, so this is filed as a proposal rather than a PR.A lesson from that attempt is worth recording: it drifted toward mass generation — helper scripts named
generate-nl-draft.jsandfill-nl-machine.jswere created to populate the dictionary from machine output. That directly contradicts what "curated" means here. If scripting is used at all, it should scaffold structure and surface gaps for human review, not fill in terminology.