AI feature calls can now be marked accepted or rejected
GET /api/usage/llm/events lists individual AI-backed calls, and PATCH /api/usage/llm/events/{id}/outcome records whether one was accepted, alongside a quality score.
The Usage page's pricing events have recorded outcome and quality since 2026-09-20. AI-backed feature calls (agent generation, task estimation) never gained the same record: the table storing them had no per-row read or write path at all, so a call's outcome was silently dropped in a different way than a pricing call's was.
GET /api/usage/llm/events now lists individual AI calls for the authenticated tenant, newest first, and PATCH /api/usage/llm/events/{id}/outcome records whether one's answer was accepted or rejected, and/or a quality score from 0 to 1. Both are optional and absent until reported: a null outcome means "not yet reported," never "rejected." The Usage page's AI tab shows the new Recent AI Calls table with the same accept and reject controls the pricing events table already has.