AcademyLens glossary packs are useful only when reviewers can find mistakes quickly. The repo includes an audit packet generator for AI review, native review, and community review.
npm run glossary:audit
npm run glossary:audit -- --locale=jaThe command writes Markdown and JSON packets to dist/glossary-audit/. The folder is ignored by git.
Each packet includes:
- every source/target pair for the locale
- source IDs used for the term
- review flags such as
single-word-overreach-risk,latin-target-review, andtarget-same-as-source - protected-term collision checks
- Check
single-word-overreach-riskterms first. Short words such asmodel,module,lesson, andexamplescan over-apply in normal course prose. - Check terms with Latin text in the target language. Some are intentional, such as
JSON mode, but others may be lazy transliteration. - Compare high-impact terms against OpenAI Academy pages and OpenAI developer documentation.
- For
llm-draftedpacks, mark issues without promoting the pack status. - For
llm-auditedpacks, confirm the second-pass audit signal and runnpm run check:glossary-quality. - Promote a pack only after native/community review or a documented review process.
llm-drafted: AI-assisted draft. Good enough for beta correction, not enough to claim native quality.llm-audited: A second AI review pass has been applied, recorded, and locked bynpm run check:glossary-quality.community-reviewed: Reviewed by a contributor with language knowledge.native-reviewed: Reviewed by a native or near-native speaker.reviewed: Final project-maintainer status after review evidence is complete; community-reviewed packs remain below this level until pending QA signals are closed.
npm run check:glossary-overreachThis script runs selected Academy/Gradual fixture text through the glossary masker and fails if glossary placeholders become suspiciously dense. It is not a substitute for manual DOM QA, but it catches the most obvious "the dictionary got too aggressive" regressions.