Actigraphy

Almost every sleep tracker that is not wired to your scalp is running some version of actigraphy. A sensor samples movement many times a second, collapses that into one activity number per epoch, and a scoring algorithm reads the pattern of stillness to decide whether you were asleep. It has decades of clinical use behind it, mostly because it can run for weeks in someone’s own bed rather than for one night in a lab.

What it is genuinely good at is timing. When you got into bed, when you fell asleep, when you got up, and how much those times drift from one day to the next. Regularity scores are built almost entirely on this kind of data, because you need many nights to compute one and nobody spends many nights in a lab.

What actigraphy misses

Stillness is not sleep. Lie in the dark for forty minutes thinking hard about tomorrow and a movement-based scorer will usually call most of it sleep, which quietly inflates total sleep time and sleep efficiency while understating wake after sleep onset. The bias runs one way, and it runs hardest in exactly the people most curious about their numbers: the ones who lie still while awake.

It also cannot see stages. Brain activity is what separates REM from deep sleep, and an accelerometer has no access to it. Stage numbers derived from motion, including the ones Tock shows on a phone-only night, are estimates. Measured stages need polysomnography, or a wearable feeding its own data into Apple Health.

Tock runs two channels overnight, accelerometer actigraphy and the microphone, and the accelerometer half is where smart wake looks for a lighter moment in the 30 minutes before your alarm. Your set time is always the fallback, which matters precisely because the underlying signal is an estimate. For the practical version of all this, including where a phone on the nightstand holds up against a wrist and where it does not, read sleep tracking without a wearable.