A user reports that the assistant gives an outdated policy. An FDE first identifies the affected workflow and request, then checks whether the source document, ingestion pipeline, retrieval filter, or generated answer caused the problem. The response depends on where evidence points.
Suppose the source was updated but old chunks remained searchable. The fix may involve version-aware ingestion and deletion rather than a new prompt. The engineer reproduces the issue, adds a regression case, validates the fix, and explains the impact and recovery to the operator.
The same day may include a discovery call, coding, a design review, and a status update. Communication should make decisions and uncertainty clear. “We are improving the AI” is less useful than explaining the failed stage, the temporary fallback, and the acceptance check for the fix.
Exercise: write a short user update for this incident and a technical investigation checklist. Keep confirmed facts separate from hypotheses.
Check: the update explains impact and next action without claiming a fix before it has been verified.