When someone files a Jira ticket about a bad translation – wrong term, off tone, broken instruction – the fix usually isn't in the translation itself. It's in the engine config. The new triage workflow opens that ticket, reads the offending output, and proposes the actual change: a glossary item to add, an instruction to tighten, a model config to adjust. You review the suggestion, accept or edit it, and the engine is fixed for every future translation – not just the one that was reported.
In parallel, the engine started scoring two built-in auto-reviews on the backend this week, ahead of the dashboard UI in W14: glossary adherence and per-instruction pass/fail. Each translation gets a pass/fail per glossary entry and per instruction, with reasoning attached when something fails.
Also shipped#
- Scorer prompts now receive the localization context that produced each translation. Scorer reasoning aligns with the engine's actual configuration instead of evaluating in a vacuum.
- Glossary edits and deletions persist correctly. Adding and deleting target translations on glossary items now saves; word counts and search no longer skip non-string values.