departures / intervals / returns
Departures below, returns above. A field of intervals, with the few retained state and care references shining through.
One strand is one eligible gap. In calendar mode, the lower endpoint sits at the departure timestamp and the upper endpoint at the return timestamp, using the same linear time scale as the dates below.
The horizontal span is the actual elapsed gap. On this multi-year view, a gap shorter than a day occupies less than a pixel and appears almost vertical. A tall strand means large activity counts in the surrounding windows; its vertical length does not represent days.
Expanded mode preserves the earlier sweeping form for closer visual inspection. It deliberately enlarges gap widths logarithmically and offsets the endpoints. Its date axis is removed and a caption identifies the enlarged geometry. Those widths cannot be read as calendar distances. The bend between endpoints is a drawing choice.
Vertical endpoints carry the selected behavior measure during the 24 hours before and after. The lower endpoint is before, the upper endpoint after. Raw height uses a linear scale for the recorded values, keeping large differences visible. The compressed option restores the previous log(1 + value) scale for counts and words. Shares always use a linear 0–100% scale. No moving average or neighboring-strand smoothing is applied; each endpoint still summarizes its own 24-hour window.
All filters and held-out selections share the same full-archive scales. An unavailable endpoint leaves an open strand. If both endpoints are unavailable, a small dot marks only the return date.
Color marks available return-day references: care takes visual precedence, then high / mixed, then low. The exact reader retains all categories. Red triangles mark high/mixed references, blue circles mark low references, and yellow diamonds mark care references. White outlined squares have no such retained reference or unavailable annotations; that does not establish a well period.
A gap is elapsed time between consecutive retained authored records. Both 24-hour windows must lie within archive bounds; these bounds do not prove exhaustive collection. The last pre-gap and first return record are included. Night share needs at least five records; lexical denominators are recalculated per side. The thresholds test sensitivity to the gap definition. Windows can overlap, so pairs are not independent. Reference coverage means existing codebook candidates have coding, not validated recall. Now/recent return-day statements can be uncertain; a gap is not measured sleep or an episode transition.
Full collection method · Nine forms