I've had some form of a fitness tracker attached to me for years now. An Oura Ring on my finger, Apple Watch on my wrist and at this stage, I've worn too many rings and bands for me to keep track. The question I've gotten asked most times, usually by a friend eyeing my sleep score with some degree of suspicion, goes something like "How does it actually know that?"

It's a valid question to ask. My ring has no idea how badly I've slept. My watch doesn't know I'm stressed about meeting my deadline. All they have is some set of numbers being produced off of a sensor every few seconds. Beyond this point, anything else, such as my sleep score, readiness score and "Your body needs recovery," comes after all sorts of mathematics going on behind the scenes. I wanted to actually break down what's going on between the raw signal and the insight you see on your phone, because once you understand it, you read your own data very differently.

The Sensor Only Sees Noise

Start with what the hardware is physically doing. For almost all wearables, there are a few specific types of sensors used: photoplethysmography (PPG) for heart rate monitoring, accelerometer for activity monitoring, gyroscope, a temperature sensor and on newer devices, a SpO2 sensor for blood oxygen.

The PPG sensor sends light through your skin and measures the intensity of light that bounces off. As the amount of light absorbed varies based on the blood flow in the capillaries at the very moment when the light passes through them, the sensor actually tracks fluctuations of the reflection intensity and labels it as your pulse.

Raw data from such a sensor is very messy on its own. Any activity, changes in the lighting, even your skin color and pressure of the band on your wrist affect the signal significantly. This way, if you have got a weirdly high heart rate during your workout, you know who to blame now.

Cleaning the Signal Before It Means Anything

Before we get a trustworthy number, the raw signal goes through an array of filters. They filter out the motion artifacts, remove the outliers that do not fit into the heartbeat pattern and correlate the accelerometer data in order to determine whether the increase in heart rate was caused by your heartbeat or by fiddling with your phone.

This is where all the real engineering wizardry happens and it is a tough signal processing problem. Most techniques use the accelerometer as the "reference signal" because the accelerometer detects only motion, the algorithm can compare it with the PPG signal and remove whatever part matches. Usually, this is achieved using adaptive noise cancellation, an approach that continually updates a filter to remove an interference pattern, just like noise-canceling headphones do.

One widely cited method pairs this with notch filtering, which strips out a specific frequency band once the true heart rate frequency has been estimated and gets the error down to under 1 beat per minute on intensive running data in lab testing. Other approaches skip the accelerometer completely and use spectral analysis of the PPG signal itself to distinguish between the heart beat frequency and the motion frequency. Of course, no technique is foolproof by itself, which is the reason why commercial devices usually use a combination of two or three techniques.

From Numbers to Meaning: Heart Rate Variability

Once you've got a clean heart rate signal, the next layer is heart rate variability or HRV.  This is the parameter that quietly calculates the majority of the "readiness" and "recovery" scores on every major wearable device out there.

It represents tiny fluctuations in time between heartbeats. More varied beats indicate a relaxed and recovered nervous system. The particular number that most wearables display is RMSSD, root mean square of successive differences between heartbeats. RMSSD is considered a time-domain metric that represents parasympathetic nervous system activity, meaning the "rest and digest" side of your nervous system rather than the "fight or flight" side of it. An elevated RMSSD number suggests an optimal condition when the body is more relaxed and recovered, whereas suppressed RMSSD numbers can occur during stressful situations, illness, or under-recovery.

According to a 14-day observational study that utilized chest-strap sensors, individuals with higher morning RMSSD showed improved sleep, less fatigue and lower stress levels despite the moderate effect size of RMSSD and considerable variance among different people. It's important to understand that the heart is not supposed to beat in a metronomically strict pattern and if it starts to do so, that's usually a sign something's off.

This is the point where we transition from raw sensor data to the information about ourselves. A figure like "62 bpm resting heart rate" doesn't really provide anything on its own. However, if a device notices a sharp decline in HRV trend over the period of several weeks, this means that your body is fighting some disease before you feel its effects. I had such cases with my Oura ring, which indicated my decreased "readiness" a full day before I started feeling unwell.

Sleep Staging Is Basically Educated Guessing, Done Well

Sleep monitoring technology relies on the same basic approach, only with more layers. Devices use heart rate, HRV, movement and sometimes body temperature and respiration rate to guess what kind of sleep stage you're in, be it light, deep, REM or awake.

The key point is that none of those devices is actually able to read your brain waves as a polysomnography test will do. The closest thing to an exception would be EEG-powered sleep wearables, which use headbands with real electrodes to capture brainwaves, the very signal that a sleep study conducted using polysomnography can measure. The difference lies in the approach taken, as opposed to other rings and wrist-worn wearables which rely on accelerometers and PPG to determine the sleep stage.

What the other, more discreet wearables like smart rings and bands do is they compare patterns. During deep sleep, there is a steady heartbeat with minimal variations and very little body movement. During REM sleep, there is high variability in heart rate and quick changes in breathing. During light sleep, there is something in between and the algorithm tries to guess what phase you are in at the moment by comparing your current data to all possible known patterns.

And they are more accurate than you might think, but not as accurate as they are commonly thought to be. In a validation study conducted in 2023, a neural network was trained on only motion and HR metrics and then tested on polysomnography scoring. The neural network scored 77.8% accuracy on epoch-wise prediction of a 4 stage (wake, light, deep and REM) model, achieving a kappa coefficient of 0.638, a statistic indicating an agreement above and beyond the chance. To explain the meaning of this number, substantial, not near-perfect, agreement is what is generally considered acceptable in the field.

A separate study on Oura Ring used 440 nights of polysomnography and ring readings from 106 participants and showed that including autonomic nervous system measurements such as HRV into a motion-only classifier significantly improved classification accuracy. This is why multisignal wearable devices have superior sleep tracking ability compared to simple step-counters, more signal types go into the very same inference process.

That is also the reason why your sleep score might be slightly different from night to night, even if you think you slept the same, time, duration and how good your sleep was. The algorithm does not measure your sleep, it makes its best guess based on external data.

Why Two People Get Different Insights From the Same Device

But here comes the tricky part. Two people wearing the exact same piece of hardware and performing the exact same exercise will have their scores vary significantly. And it is supposed to be like that. This is where the personalization layer does its job.

Most modern-day devices collect data on your individual baseline, which consists of your resting heart rate, your usual HRV readings, your temperature and everything else that makes your biometrics unique. This is not a vague “it learns you” approach but a concrete rolling comparison. For example, Oura has documented that it gives more weight to a 14-day HRV average based on your recent nights before comparing it with a more prolonged 3-month average and detecting any deviations from your regular pattern. This way, each time you perform an activity, its effect on your biometrics gets compared to your own norm and not some general one.

This is a big part of why the female health tracking conversation has gotten so much more serious in the last few years. Each stage of the cycle affects resting heart rate, body temperature and HRV differently and a system which compares your readings to a general baseline will mistake the natural fluctuation for a sign of stress and insufficient recovery. This is why most of the good trackers start incorporating the cycle into their model to avoid such mistakes.

Where the Insight Layer Still Falls Short

I will be transparent about the limitations in this regard, as I believe there is an excessive trust in these scores a bit too blindly sometimes.

The optical sensors still have trouble with darker skin and low blood perfusion and the accuracy truly varies from person to person. This isn't a minor caveat, it’s a well-documented and persistent accuracy gap. A 2024 systematic review and meta-analysis of 23 pulse oximetry studies and 4 wearable pulse-rate studies, which included 197,353 pairs of measurements, showed that SpO2 and pulse rate readings can breach FDA and industry standards of accuracy for all races of skin pigmentation levels, with more pronounced bias being observed in patients with darker skin.

An illness, alcohol consumption, long flights and a restrictive band on the arm will negatively impact the readiness score without impacting your physical health in any way.

A poor night's sleep in isolation does not mean anything but the pattern over weeks is what carries the message, not a single score.

I have learned to view these values as a guide more than a diagnosis. If my readiness score is poor for one single day, I do not get alarmed. It is the consistent downward trend over a period of two weeks that gets my attention.

Temperature and SpO2 Add Another Layer

Heart rate and activity receive all the glory, but skin temperature and SpO2 are quietly doing a lot of work too. Temperature sensors detect minor fluctuations around the baseline temperature. Even a sudden increase at night is one of the first indicators your body is struggling with something, long before any symptoms appear during the daytime. It's also the backbone of a lot of cycle tracking features, since basal body temperature shifts predictably across a cycle if you know what to look for.

SpO2 measurement helps to get an idea about breathing quality when you are sleeping. It does not replace proper medical examination of your breathing problems, such as sleep apnea, but continuous drops may prompt you to visit the doctor. All those parameters alone don’t give you a lot. Combined and evaluated through the lens of your own history, they tell a story worth listening to.

Why the Same Raw Data Can Tell Different Stories

This is a question that often pops up while doing reviews of different wearables. The thing is, two different pieces of technology can interpret the same data in their own way and provide completely different numbers, since the interpretation layer is proprietary.

A 2025 academic review analyzed 14 such composite scores offered by 10 leading manufacturers of wearable health devices, based on publicly available technical documentation and user manuals rather than speculating about the black box inside. In 86% of the scores, HRV was used, while the resting heart rate was incorporated in 79% of the scores and physical activity as well as sleep duration were included in 71% of them. However, the review also revealed some structural discrepancies in the way each company employs these factors.

The documentation for Oura readiness score reveals that it compares your HRV value to its average during both 14 days and 3 months prior to the current measurement and incorporates your body temperature deviation, which is unique for Oura compared to other major manufacturers. The recovery score provided by WHOOP uses only one HRV value measured during your deep sleep phase, specifically, slow-wave sleep. This choice of approach was deliberate in order to avoid the noise caused by caffeine or screen consumption and any potential stress prior to bedtime.

Lastly, the Body Battery score provided by Garmin represents neither a one-nor two-day snapshot, but an ever-changing indicator of energy reserves, based on the proprietary Firstbeat analytics engine that drains throughout the day depending on your activities and stresses and then recharges during sleep. It is for this reason that it may lag behind overnight recovery that WHOOP or Oura can already detect on the morning after.

This is also why chasing a perfect score across multiple apps at once is kind of a losing game. You need to pick one system and get used to the way it interprets your health parameters, letting it collect your history.

It's the Trend, Not the Number

What actually makes a fitness tracker useful isn't any single number on any single day. What really matters is having several months worth of data, which are slowly compiling an image of what is "normal" for you. A resting heart rate of 58 is not really impressive by itself. But a resting heart rate of 58 when you normally have 50 is definitely worth investigating.

The raw data from the sensors is actually the least impressive part of this entire process. The most important thing about it is what happens after: filtering, establishing a baseline and detecting patterns based on your history of data. It is what makes a photodiode attached to your finger capable of figuring out that your body is under recovered before you've even noticed it yourself.