When a completed Whisper transcript already has a successful AI analysis and a later speaker-diarization rerun succeeds with more distinct speakers than the previous diarization result, Tools automatically regenerates the analysis.
The regenerated analysis uses the speaker-aware transcript, including generic speaker labels and timestamps when available, instead of the plain transcript text. Existing analysis language preferences are preserved. If translated analysis variants already exist, the normal analysis completion path refreshes those variants from the regenerated analysis.
No automatic reanalysis is started when diarization fails, when the speaker count is unchanged or lower, or when the transcript did not already have a completed analysis. This prevents retry loops and avoids enabling AI analysis for jobs where analysis was not previously used.
If automatic reanalysis fails after diarization itself succeeded, the completed transcript and diarization result remain intact. The failure is reported through the normal always-on Tools operational diagnostics path without transcript content or credentials.
Implementation issue: #1587.