Hi, We are integrating HKQuantityTypeIdentifierTimeInDaylight into a research application and have a few questions about how developers should interpret the data returned by HealthKit. Specifically: Should TimeInDaylight samples be treated as having a fixed minimum temporal granularity (for example, approximately 5-minute intervals), or is the sample duration implementation-dependent and subject to change? Is there any expected latency between a daylight exposure event and the corresponding TimeInDaylight sample becoming available through HealthKit? For example, are samples intended to appear shortly after exposure, or only after periodic processing and synchronization? In the Health app, each Time in Daylight sample displays a Maximum Light Intensity (lux). Is this value available through the public HealthKit API (e.g., metadata), or is it only used internally by the Health app? More generally, should developers consider TimeInDaylight to be a high-level derived metric rather than expecting a direct correspondence with underlying ambient light sensor observations? Thank you.

Dear SensorKit team, We are currently working with SensorKit ambient light data under our approved SensorKit entitlement for research use. We would be grateful for some technical clarification on the sampling strategy for the SRSensor.ambientLightSensor stream, as this directly affects how we analyse and report the data. In exported data from Apple Watch, we observe that ambient light samples do not appear to follow a fixed sampling cadence. Instead, the data appear burst-like: in one short window, we see many samples with inter-sample intervals around 100 ms, occasional near-duplicate timestamps, and then gaps of around 10 to 30 seconds with no samples. This suggests that the stream may be adaptive, event-triggered, buffered, or subject to system-level sampling decisions. We also noticed a related discrepancy when comparing the SensorKit ambient light trace with the Health app display for a corresponding Time in Daylight sample. In one example, the Health app shows a 5-minute Time in Daylight interval with a “Maximum Light Intensity” value of 9,493 lux. In the SensorKit ambient light trace around that period, the raw samples show a different maximum depending on the precise time window considered, including higher values shortly before the HealthKit interval start and lower values within the subset of SensorKit samples we inspected. We realise that Time in Daylight may be generated by a separate internal aggregation or classification pipeline, but this comparison raised the question of how closely SensorKit ambient light samples should be expected to correspond to the light intensity values displayed in Health. Could you clarify the following points for SensorKit ambient light data? Is SRSensor.ambientLightSensor sampled at a fixed cadence, or is sampling adaptive / event-driven? If sampling is adaptive, what factors influence sampling density? For example, changes in illuminance, device or wrist motion, device orientation, display state, app state, power state, charging state, or other system-level conditions. Are ambient light readings buffered and delivered or exported in bursts? Do SensorKit timestamps correspond to the physical sensor acquisition time, processing time, or the time at which the sample is made available through SensorKit? Are duplicate or near-duplicate ambient light samples expected in SensorKit exports? Are there circumstances under which ambient light sampling is suspended, downsampled, or suppressed? Is SensorKit ambient light expected to match, approximate, or differ from the light intensity values shown in Health app Time in Daylight sample details? Is the “Maximum Light Intensity” shown for Time in Daylight computed from the same underlying ambient light sensor stream exposed through SensorKit, or from a separate internal stream or aggregation? Are there recommended practices for analysing SensorKit ambient light data, especially with respect to irregular sampling, burst sampling, missing intervals, and aggregation to longer time windows? Is the sampling strategy the same across Apple Watch hardware versions, or should researchers expect device-specific differences? We do not need proprietary implementation details. Our goal is to understand the methodological constraints well enough to analyse the data appropriately and describe the limitations accurately in scientific work. Thank you!