Every dead-air incident has the same shape. Something was wrong with the schedule, nobody looked, and the problem surfaced at the worst possible time.
The fix is not more discipline. It is having the software check continuously and tell you.
What gets checked
A background, debounced scan runs over the loaded playlist looking for:
- Missing files — a path that no longer resolves. The classic cause of dead air, usually a moved or renamed folder.
- Missing cue points — tracks that will mix on their fade-out tail rather than a marker.
- No loudness measurement — tracks with no cached analysis, which will sit at whatever level they were mastered at.
- Weak rotation category matches — items the generator was not confident about.
One chip, three states
The result is a single indicator on the status bar: Ready, Review or Risk. Click it and a window lists every issue in detail.
Three states is the right number. A count of warnings gets ignored once it is never zero; a binary good/bad has nowhere to put "playable but not ideal". Ready means go, Review means look when you can, Risk means look now.
When it runs
The scan triggers on playlist refresh, on loudness cache updates, and when metadata batches complete — the moments when the answer could actually have changed. It is debounced, so a burst of edits produces one analysis rather than fifty.
System health, separately
A second indicator covers the machine rather than the schedule: licence validity and expiry proximity, audio engine output availability including recent faults, SQLite connectivity, and AI configuration state.
Informational rows deliberately do not escalate the overall status. AI not being configured is worth showing but is not a reason to display "attention needed" on a station that does not use it.
It refreshes on the events that matter — audio device changes, completed AI calls, preferences closing — with a two-minute timer as a slow fallback for time-only drift such as a licence day countdown.