Every benchmark comes from somewhere. Here is where.
These numbers are in the app, on the cards, with their source written underneath. They are not there to impress: they are there so you can go and read what they rest on, and decide for yourself what to make of them.
Your body index
Example
You are within the reference range. It's a population benchmark: it doesn't tell muscle from fat.
World Health Organization reference range for adults: 18.5 to 25.
7,000
steps a day
The benchmark large cohort studies point to. A benchmark, not a promise about you.
Ding et al. (2025), Lancet Public Health, 57 studies
7h
of sleep
Your nights are read against this benchmark, and against your own weeks.
Yin et al. (2017), J Am Heart Assoc
208 − 0.7 × age
your maximum heart rate
And not "220 minus age", which underestimates it past 40.
Tanaka et al. (2001), 351 studies, 18,712 subjects
46 studies
on resting heart rate
Your resting heart is read against that base, and against your own weeks, not against a round number.
Zhang et al. (2015), CMAJ, 1.2 million participants
0 to 3
reps in reserve
Going to failure every time adds nothing, and tires you markedly more.
Robinson et al. (2024), Sports Medicine
The night, beyond its length.
Duration remains the first benchmark: it is what the consensus statements cover. For the rest of the night, here is what is published, and how far each figure holds.
85%
sleep efficiency, at least
Time asleep over time spent in bed. With it, 20 minutes awake at most after falling asleep and at most one awakening longer than five minutes, two after 65: the continuity benchmarks, where the consensus is clear.
Ohayon et al. (2017), Sleep Health, National Sleep Foundation consensus panel, 277 studies · doi:10.1016/j.sleh.2016.11.006
16 to 20%
deep sleep, in the laboratory
The same panel’s range for adults aged 26 to 64, with 21 to 30% REM. The consensus there is weak: the app shows it as a laboratory observation, never as a target.
Ohayon et al. (2017), Sleep Health · doi:10.1016/j.sleh.2016.11.006
43 min
of deep sleep not seen by an Apple Watch Series 8 (one study)
Against polysomnography, over one night in the laboratory. On a Series 7, at home, another study finds no clear gap in deep sleep, but 74 minutes less light sleep and 30 minutes more REM. Hence the app’s rule: a watch share is not read literally against a laboratory range.
Robbins et al. (2024), Sensors, 35 adults; Alhejaili et al. (2026), Sleep Medicine, 19 Apple Watch wearers · doi:10.3390/s24206532 · doi:10.1016/j.sleep.2026.109059
90 min
per cycle on average, with wide gaps from one cycle to the next
The average length of a cycle in 205 healthy adults. In adults, no meta-analysis sets either the length or the number of cycles: the app counts them from the stages the watch estimates, to describe the night, not to score it.
Rosenblum et al. (2025), eLife, 205 adults · doi:10.7554/eLife.96784
When the literature does not settle a question, the app does not settle it either. What we will not tell you.