Added
- Analyze tone history shows dispatch transcripts — each suggested tone set lists the post-tone dispatch voice from the calls that produced it, reusing the live tone-alerting voice attachment (same-call voice, or the next voice call on the talkgroup within the pending-tone window).
Changed
- Auto-learn label uses the resolved post-tone voice and a stronger OpenAI prompt that locks onto the single station/unit consistent across transcripts and ignores one-off addresses, patient details, names, and times.
- Auto-learn frequency tolerance default 10 → 20 Hz (candidates within 20 Hz of an existing set are treated as the same set).
- Auto-learn minimum A/B separation 90 → 40 Hz so tight legitimate pairs can be learned; detection/alerting of stored sets is unaffected.
- Tone set editor cleanup — descriptors under inputs, removed unused "Sequence Min Duration", full-width layout, reduced input height, corrected tolerance hint.
Fixed
- Unit edit could not be saved — editing a unit's label/ID now enables Save immediately (live commit) instead of only after collapsing the row.