Running Speed

Running speed describes distance travelled per unit of time. HealthKit stores a discrete speed quantity; PulsHealth uses metres per second. Apple's HealthKit reference documents automatic outdoor-running collection on Apple Watch. Pace is a useful alternative display, but it is calculated from speed rather than stored as this type.

The July–September 2026 example contains 11,707 readings from one source group across 13 runs. These are observed values, not population reference ranges or training targets.

The histogram weights every stored reading equally, so longer or more densely recorded runs contribute more. The second panel gives each paired run one point. All extremes remain visible; an outlier flag is a reason to inspect its timestamps and context, rather than proof of an error or a clinical abnormality.

The median positive gap between sample starts in this quarter was 2.57 seconds. This calculation includes gaps between runs; the mean would be dominated by those gaps. All quarter readings were point records with equal start and end timestamps. That is the stored shape, not proof that the sensors only operated once every 2.57 seconds.

Earlier months contained longer interval records. Apple documents that workout samples may be condensed or coalesced, so a change in sample counts or interval length is not automatically a change in training volume. Recording cadence is also different from running cadence in steps per minute.

Speed decreased by a median 0.26 m/s from the first to last elapsed-time third across 13 runs. The calculation compares medians of point readings in the first and last thirds of elapsed workout time, with at least five readings per third. Pace, terrain and pauses were not controlled. It cannot establish fatigue, fitness improvement or form breakdown.

Every quarter reading overlapped a recorded workout; none overlapped a recorded asleep interval. The example therefore supports comparisons within and between runs. A day without readings means no stored observation of this type, rather than a physiological zero.

For a positive speed, minutes per kilometre = 1000 ÷ (60 × speed in m/s). The example median of 2.55 m/s is about 6:33 min/km. A speed of 3 m/s corresponds to about 5:33 min/km. At zero speed the conversion is undefined; retain the zero in speed analysis rather than inventing a finite pace.

Pace is an inverse transformation, so averaging pace readings and converting average speed can produce different answers. A workout's moving-time pace also requires decisions about pauses and distance. The distribution here describes stored speed readings, not the official workout average.

Across 13 paired runs, faster run-median speed was associated with longer single-step stride length (0.83), higher running power (0.88) and shorter ground contact time (−0.71, Spearman correlations). These are workout-level comparisons, not separate independent experiments for every sample. Route, grade, workout design and conditions can change all these metrics together.

HKQuantityTypeIdentifierRunningSpeed is a discrete quantity type, introduced in iOS 16/watchOS 9. Use averages, distributions or matched comparisons rather than summing readings. Automatic outdoor-running support is documented by Apple; HealthKit's speed page does not specify a model cutoff.

Raw quantity samples and workout-linked series are kept separate. This quarter contains 11,707 raw readings and 11,707 series points for this type; these counts are representations, not additional independent measurements to pool together. Source revision groups are provenance, not proof of a permanent device identity.

The research uses an October 2025–September 2026 extract with a July–September focus in America/Los_Angeles. Workout relationships use fully contained readings, the same source group and uniquely overlapping running workouts, with one median per run. Sleep and general workout context use timestamp overlap rather than explicit HealthKit associations. Context beginning before the extract boundary may be missing. The private executed notebook retains all eleven analysis facets, source changes, outlier checks and unsupported questions. Public figures omit dates, source names, identifiers and routes.