Open any translation in the dashboard and you can now see, line by line, which glossary term passed, which failed, and why. Same for every instruction the engine carries. Until this week, "the glossary wasn't applied correctly" was a sentence in a Jira ticket; now it's a row with a verdict and the model's reasoning. The auto-reviews running on the backend since W13 finally have a UI – pass/fail per glossary entry, pass/fail per instruction, reasoning on each failure.