Audits, glossary and rules

Ask how a language is doing, turn the findings into fixes and make every future translation better.

Three things shape every translation besides the source text: the context of the string, the glossary and the rules. Audits are how you find out what to put in them.

Glossary

Glossary terms are brand and product names that must never be translated, and terms with a required translation per locale. Every translation prompt includes them.

npx @meterapp/glossa upsert_glossary_term --json '{"term":"Meter","do_not_translate":true}'npx @meterapp/glossa upsert_glossary_term --json '{"term":"workspace","translations":{"es":"espacio de trabajo","de":"Arbeitsbereich"}}'

Rules

Rules are standing instructions appended to every translation prompt, for one locale or all of them: tone, formality, length limits, how to treat units or dates.

npx @meterapp/glossa add_rule --json '{"locale":"es","body":"Use tú, never usted."}'npx @meterapp/glossa add_rule --json '{"locale":null,"body":"Keep button labels under 20 characters."}'

Better context fixes the cause rather than the symptom: update_unit with a description of where a string appears, and retranslate: true, redoes its machine translations in every locale with the new context.

Audits

An audit reviews one locale as a whole and answers a question you ask, such as "How is our Spanish?" or "Are brand names consistent?". It samples translations, scores the locale, answers the question, and files tasks:

  • audit_finding: a specific translation that is wrong, with a proposed fix. apply sets it.
  • glossary_suggestion: a term to fix or protect.
  • rule_suggestion: an instruction that would prevent the same mistake next time.
npx @meterapp/glossa run_audit --locale es --question "Are we consistent about tú and usted?"npx @meterapp/glossa get_audit --json '{"audit":"<audit_id>"}'npx @meterapp/glossa tasks --kind audit_finding

Audits run as background jobs; run_audit returns an audit_id and a job_id. Read the result with get_audit once the job is done.

The loop

  1. Audit a locale with a pointed question.
  2. Apply or edit the findings that are right; dismiss the rest.
  3. Add the rules and glossary terms that explain the findings.
  4. Audit again later. The same mistakes should not come back, because every new translation now follows the rules.