Edu-Carone-SA gets its legal-AI operations room working
A QA repair pass fixes the gaps that made a live production run hard to trust or follow.
The biggest fix is basic but consequential: production runs now return written answers, while benchmark testing keeps its existing scoring behaviour.
- Production exports retain the refined question and its flow data.
- The dashboard stops waiting indefinitely on first load and can show aggregate token use and cost.
- Live operations record each question when it happens, with real timestamps, accurate progress, feed scrolling and a short hand-off to the after-action report.
- Pricing and usage pages keep readable colours while their styling loads.
This is operational polish, but it matters when a legal team needs to inspect a run rather than guess what happened.
So what Teams evaluating AI-assisted legal workflows should care because the fork makes production activity, costs and audit-ready outputs easier to see in the moment.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?