Useful estimates, with their limits visible.

Boop distinguishes decoded strap observations from values it estimates. Derived Apple Health writes are labeled as estimates and versioned with the Boop algorithm revision where Health metadata supports it. The revision belongs to the installed build and may differ across updates.

Boop is not a medical device. Outputs may be approximate and must not be used for diagnosis or treatment.

Boop calculates time-domain variability from plausible beat-to-beat intervals and rejects obvious outliers before deriving an SDNN estimate. Definitions and cautious interpretation follow the 1996 ESC/NASPE Task Force standards. Wrist-derived intervals are not equivalent to a diagnostic ECG.

Sleep/wake estimates combine wrist activity and available physiological signals in fixed windows, informed by the public actigraphy approach in Cole, Kripke, and colleagues (1992). Recovery-oriented summaries use personal rolling baselines rather than population diagnoses. Missing or noisy data lowers confidence.

Energy estimates combine demographic inputs, heart rate, and detected activity. The implementation is informed by the public prediction work of Keytel and colleagues (2005) and the reevaluated Harris–Benedict equations. Predictive equations have wide individual error and are wellness estimates, not measured calorimetry.

Motion artifacts, fit, skin contact, firmware changes, missing samples, physiology, and algorithm assumptions can all affect results. Boop’s outputs have not been validated for diagnosis or treatment. Algorithm versions may change, so values from different versions should not be assumed interchangeable.