Structured workouts vs. device laps

When you ask for the intervals of an activity, the rows you get back can mean two completely different things — and telling them apart matters.

Two kinds of “interval”

An interval in a structured workout is a planned step: “5 minutes at threshold”, “8 x 30 seconds”. A device lap is something your bike computer or watch created on its own — most often an automatic split every 1 km or 1 mile.

Both arrive as interval rows from get_activity_intervals, and from the numbers alone they look alike. An AI assistant that cannot tell them apart will happily report that you “missed interval 3 of your workout” when interval 3 was just the third kilometre of a steady ride.

How icuvisor labels the source

To head off that guess, the response carries additive metadata:

• _meta.interval_source — structured_workout, device_laps, manual_added, mixed, or unknown.
• _meta.auto_lap_suspected — true when the rows are generic and near-uniform (1 km, 1 mi, or another supported fixed duration), the signature of device auto-laps.
• _meta.interval_source_caveat — present when the interval rows need extra caution, such as a single averaged row that may be an imported or collapsed lap.
• _meta.recommended_next_tool — present with the caveat when a follow-up tool can provide better evidence, currently compute_activity_segment_stats for stream-derived segment analysis.

What it means for analysis

When auto-laps are suspected, the rows are splits, not planned segments. An analysis should describe them as distance or time splits and must not claim the athlete hit or missed individual workout steps — there were no planned steps to hit. When the source is a genuine structured workout, step-by-step compliance is fair game.

A single unknown interval row is also not proof that the athlete did one steady block or skipped all reps. Some device/import paths can collapse a sprint or anaerobic workout into one averaged lap. In that case icuvisor exposes the caveat and routes the assistant to compute_activity_segment_stats over explicit time or distance bounds. Segment stats can test likely work windows from streams, but they still do not reconstruct exact reps from one averaged lap by themselves.

This is the same terse, honest-by-default stance behind the rest of icuvisor: surface what the data actually supports, and signal uncertainty rather than paper over it. The activity retrospective recipe leans on this when it breaks a session down rep by rep.