Let the curve
tell you what it can.

The selected synthetic day contains 97 quarter-hour timestamps from 00:00 through 24:00 UTC, with five deliberately missing readings. It does not use uploaded or personal data. Rates use adjacent readings only, divided by 15 minutes. “Steady” means an absolute rate below 0.25 mg/dL/min, a UI convention rather than a clinical rule. Fastest rise/fall are extrema of this adjacent-interval rate. Highest/lowest are observed values, not interpolated extrema.

Coverage counts a 15-minute interval only when both endpoints are present. Five missing readings leave six unusable intervals (90 minutes). Time in each display band assigns each covered interval to its starting reading; values exactly at the bounds count within range. The average is weighted by these covered intervals. Display bands are adjustable examples, not personal treatment targets. No interpolation across missing data is used. Periods are half-open for extrema and include their endpoint for interval calculations. All timestamps use UTC in this example; a real import would need its time zone confirmed.

Experimental curve-defined segmentation is an alternative to clock-time periods. Every usable adjacent interval is classified as rising at or above the selected positive slope threshold, falling at or below its negative, or steady strictly between. Default: 0.25 mg/dL/min; this is a demo convention, not a medical classification. Consecutive same-class intervals form a run. No minimum-duration filter, smoothing, or merging is applied: a single 15-minute interval can be its own segment. This preserves brief changes but can fragment noisy data. Missing endpoints or intervals longer than 15 minutes force a gap boundary, never a bridge. Segment change is last minus first glucose; average rate is change divided by observed minutes. Each valid interval belongs to exactly one segment; gap time is excluded and totals reconcile to the day. A selected segment’s historical comparison uses the same clock window, not curve-shape matching or clustering. No meals, sleep, insulin actions, or causes are inferred.

The typical-day comparison uses 28 reproducible generated prior days, never the selected day. Choose the preceding 7, 14, or 28 calendar dates. Each historical day is summarized once per 15-minute clock-time bin before taking the median and 25th/75th percentiles across distinct days. The shaded band is the historical middle 50%, not a confidence interval or treatment range. See the comparison’s calculation notes for coverage, exclusions, and difference denominators. No private uploads are used.