A fitness tracker measures far less than its display suggests. Almost every figure on the screen is an inference built from two or three raw signals, and the inference is where devices diverge.
The sensors underneath are simple
Most wrist wearables carry an accelerometer, which detects movement in three directions, and an optical sensor that shines light into the skin and reads what comes back.
The optical sensor works because blood absorbs light, so the reflected signal pulses as the vessels fill and empty. That pulsing is the heart rate the device reports.
Everything else, including steps, distance, energy burned and sleep stages, is calculated from those two streams plus whatever the user has entered about height, weight and age.
A step is a pattern the software decides to accept
The accelerometer sees a continuous wobble, not a series of discrete steps. Software looks for a repeating acceleration signature that matches walking and counts each occurrence.
That threshold has to be set somewhere. Set it loosely and hand gestures, driving over rough roads and applauding all register; set it tightly and slow strolling with a hand in a pocket disappears.
Different manufacturers place the threshold differently, which is the main reason two devices worn on the same wrist can end a day thousands of steps apart.
Energy expenditure is a model, not a measurement
No wrist device measures the energy a body uses. Measuring that directly requires analysing exhaled gases, which is laboratory equipment rather than consumer hardware.
Instead the device estimates a resting rate from the profile it has been given, then adds an activity component derived from movement intensity and heart rate.
Because the resting portion dominates the daily total and rests on population averages, the number can be confidently wrong for any particular person while still tracking that person's changes reasonably well.
Sleep staging is inferred from stillness and pulse
Clinical sleep staging reads brain activity, eye movement and muscle tone. A wrist device has none of those, so it works from movement and beat-to-beat variation in heart rate.
Those signals do separate sleep from wakefulness reasonably well, because a sleeping body moves in a distinctive way and heart rhythm changes across the night.
Splitting sleep into named stages from the same two signals is a much harder inference, which is why stage breakdowns vary widely between brands and between nights.
Consistency matters more than accuracy
If a device applies the same model every day, its errors are largely the same size every day, so the direction of change over weeks is meaningful even when the absolute figures are not.
That is the useful reading of the data. A rising weekly step average says something real; a specific calorie total says considerably less than its precision implies.
It also explains why comparing figures with someone using a different brand is close to meaningless. The two devices are not measuring the same thing, only estimating it by different rules.