Sleep trackers report a breakdown of the night into stages, presented with a precision the underlying measurement does not support. The difference between what is measured and what is reported explains where these devices are useful.
The reference method measures the brain
Clinical sleep staging uses electrodes recording brain activity, eye movement and muscle tone, scored in short epochs against defined criteria.
Those three signals are what the stage definitions are built on, and REM in particular is identified by the combination of active brain, moving eyes and absent muscle tone.
A wrist device records none of these, so it is estimating a category defined by signals it cannot access.
What a wearable actually records
Accelerometers capture movement, which distinguishes wake from sleep reasonably well because people move more when awake.
Optical sensors measure pulse at the skin, from which heart rate and beat-to-beat variability are derived, and some devices add skin temperature or respiratory rate.
These signals do change with sleep stage — heart rate and its variability differ between deep sleep and REM — which is what makes estimation possible at all.
Models translate signals into stages
Manufacturers train algorithms on data from people monitored simultaneously by wearable and clinical equipment, learning which patterns correspond to which stages.
Performance is consistently better at separating sleep from wake than at distinguishing among the sleep stages themselves.
Because the models are proprietary and updated without notice, the same night's data can be staged differently after a software update.
Systematic errors are predictable
Devices tend to overestimate total sleep in people who lie still while awake, since stillness is read as sleep.
They perform less well in people with disturbed sleep, which is the population most interested in the measurement.
Scores combining several metrics into a single number add another layer of undisclosed weighting on top of estimates that were already indirect.
Where the data earns its place
Bedtime, wake time and total duration are recorded well, and consistency of timing is among the more useful things a person can track.
Trends over weeks carry more information than any single night, since the errors are more consistent than the night-to-night variation they are measuring.
Anxiety about achieving good scores can itself disturb sleep, and a suspected disorder such as apnoea needs a clinical study rather than a wearable summary.