How Wearable Sensors Infer What They Cannot Measure

A wearable reports heart rate, sleep stages, stress and energy expenditure. It directly measures none of those, and the distance between sensor and readout explains most of its limitations.

The sensors detect light and movement

The optical sensor shines light into the skin and measures how much returns. Blood absorbs light, so the returning signal fluctuates as blood volume changes with each heartbeat.

An accelerometer records movement along three axes, producing a continuous stream of motion data. Some devices add skin temperature and electrical conductance of the skin.

Everything the device reports is derived from these streams. There is no separate sensor for sleep stage, stress or recovery, and each of those is a computed interpretation.

Heart rate comes from peak detection

Converting the optical signal into a heart rate means identifying the peaks corresponding to beats and calculating the interval between them.

Motion interferes directly, because movement changes how the sensor sits against skin and alters blood flow in the limb. Algorithms filter using the accelerometer data to compensate.

Accuracy is therefore best at rest and degrades during vigorous or irregular activity. Skin tone, tattoos, ambient light and fit all affect the returning signal as well.

Sleep staging is a classification model

Laboratory sleep staging uses brain electrical activity, eye movement and muscle tone, none of which a wrist device records. Wearable staging is inference from proxies.

The model uses movement, heart rate and beat-to-beat variation, which do differ across sleep stages, and assigns each interval to a stage based on patterns learned from labelled data.

Distinguishing sleep from wake works reasonably well because movement differs sharply. Separating the stages within sleep is substantially harder, and manufacturers use different models that disagree.

Composite scores are proprietary

Readiness, recovery and stress scores combine several derived signals into a single number using weightings the manufacturer does not publish.

Because the formula is undisclosed and revised through software updates, a score is not comparable across devices and not necessarily comparable to the same device a year earlier.

The scores are also normalised against the individual's own history, so the same physiological state can produce different scores depending on preceding weeks.

Trends survive the noise better than values

The practical consequence is that a single day's number carries limited information, while a consistent direction over weeks is more likely to reflect something real.

This is why device makers emphasise baselines and deviations. It is a reasonable response to measurement error that would otherwise dominate day-to-day comparison.

Wearables are not diagnostic instruments and are not regulated as such, except for specific cleared features. Readings that suggest a problem are a reason to seek clinical assessment, not a finding in themselves.