See where candidate patterns appeared during the night.

A count cannot show timing

Two sessions can produce the same nightly product index while their candidate events appear at different times. The timeline preserves that time context.

This can help you compare your own nights and recording conditions. It cannot identify sleep stages, oxygen changes, arousals, or clinical event severity.

What every line in the timeline gives you

For each night, the app lists candidate events produced by its acoustic model:

• Time — model-estimated start and end.
• Duration — estimated span of the candidate event.
• Type — the acoustic label assigned by the production model.
• Audio — an available local clip you can inspect.

A density visualization shows whether model output was spread across the session or concentrated in a particular period. Product color bands are visual navigation aids, not clinical AHI severity categories.

What this isn't

• Not sleep stage analysis. We don't claim to detect REM vs deep vs light sleep. Nick's "REM-clustered" pattern was inferred from time-of-night, not from EEG. Real staging needs a head sensor.
• Not a diagnosis. "80% of events between 3:30 and 5 a.m." is a pattern. What it means medically is for a clinician. We surface the pattern. We don't interpret it.
• Not independently event-validated. The published breathing benchmark evaluates 200-second windows. It does not establish the accuracy of every event boundary, duration, or label shown here.