Sleep² and Polar H10: Tracking Sleep With a Chest Strap

People who want to track their sleep tend to sit between two extremes: a consumer device (ring, watch or band) that gives a convenient score, or a polysomnography in a lab, which is the clinical reference but is not something you repeat every night at home. The sleep² app, from the Austrian developer nukkuaa, proposes a middle ground: wear a Polar sensor overnight and let an algorithm estimate your sleep stages from heart rate and its variability (HRV).

This article explains how that approach works, what scientific backing it has and what it lacks, what hardware you need, what it costs to keep up and who it is worth it for. One caveat up front: neither sleep² nor Polar sensors are medical devices for diagnosing sleep disorders. If you suspect sleep apnea, chronic insomnia or another problem, the way to go is a doctor, not an app.

How it works: from heartbeat to sleep stages

Wrist and finger devices estimate sleep by combining movement with an optical sensor (PPG) that detects the pulse. The Polar H10 chest strap, by contrast, records the electrical activity of the heart at the chest, an ECG-based method, and the intervals between beats come from that. The algorithm works on that series of intervals.

The underlying idea is physiological: heart rate variability changes with the state of the autonomic nervous system, and that state changes across sleep stages. Polar sums it up in its case study with sleep² by saying that HRV is strongly correlated with sleep stages. That is the explanation of the manufacturer and its partner; what matters is whether the result holds up when compared with a polysomnography, which is what we review next.

A nuance that often gets lost in commercial messaging: sleep² works with heart rate and HRV, not with the full ECG waveform. The app and Polar listings talk about those two data points, and we have not found documentation describing a morphological analysis of the electrocardiogram. The advantage of the chest strap is the reliability of the starting signal, not that the app performs a "medical ECG".

What the evidence says

The main reference is a study by Topalidis and colleagues published in 2023 in the journal Sensors. The relevant data:

• The model was trained on 8,898 nights of polysomnography and, with cross-validation, reached 82.1% accuracy across four classes (wake, light sleep, deep sleep and REM), with a kappa of 0.733.
• In a set of 102 healthy participants (mean age 41) with simultaneous polysomnography and sensor, the Polar H10 reached 80.3% accuracy (kappa 0.69) and the Verity Sense 81% (kappa 0.69).
• The paper describes the nukkuaa app as the one used in a sleep training study and calls it a late beta version; the name "sleep²" does not appear in the text.

These figures need to be read with care. Two of the authors have ties to the industry: one is a co-founder of nukkuaa and another works at an institute that has adopted the model in a commercial system. The authors themselves note that uncontrolled factors (tobacco, alcohol, physical activity, medication) can affect HRV and that they only modeled age and sex. And it is a single set of 102 healthy people: we have not found an independent replication specific to the H10.

Polar's page also mentions a 2025 study from the University of Salzburg, attributed to Topalidis and colleagues (the same first author as the 2023 one, so it is not independent evidence): with the Verity Sense and 18 participants (90 nights, eight systems compared), sleep² reportedly reached 83.7% accuracy across four stages and about six minutes of deviation in total sleep time. We have not been able to locate the publication, so the figure stands only as Polar states it. On the same page, nukkuaa's director says his team sets as a goal for the AI to reach 95% against the gold standard; that is a goal, not a result, and we do not use it as data.

How it compares with rings and watches

A 2022 validation with 53 healthy young adults compared six consumer devices against polysomnography. In stage-by-stage accuracy, values fell between 50% and 65% (kappa between 0.20 and 0.52), and the authors conclude that all of them need to improve at estimating specific stages. In wake detection, sensitivity ranged from 26% to 57%, and the authors consider all six devices valid for total sleep time.

This comparison has limits worth seeing: the devices were specific models from that time (among them an Oura Gen 2 and a Whoop 3.0, not current generations such as the Oura Ring 4 or the Whoop 5.0), the sample is 53 young people over a single lab night, and the two studies use different metrics and populations. They are not independent pieces of evidence that can be added together, nor do they allow us to say that one approach "beats" the other in general. The only thing both papers allow us to state is that sleep stage estimation is imperfect, whatever the method.

Hardware: Polar H10 or Verity Sense

sleep² supports two Polar sensors:

• Polar H10. Chest strap with ECG. It is the ECG-based sensor that the 2023 study evaluated. The app listing says that a Polar sensor (H10 or Verity Sense) is the most accurate option, and a reply from the developer states that the H10's raw data allows an analysis "much more precise than many consumer wearables"; that is a manufacturer claim, and it does not prefer the H10 over the Verity Sense. It is worn on an elastic strap over the sternum.
• Polar Verity Sense. Optical sensor worn on the arm. In the 2023 study, accuracy was practically the same as the H10 (81% versus 80.3%), a difference that does not allow picking a winner. If sleeping with a strap on your chest is uncomfortable, the study gives no reason to rule this sensor out.

In practice, the choice between the two depends mostly on comfort, because that determines whether you wear it every night. An excellent sensor that stays in a drawer measures nothing.

What it really costs

• The sensor is a one-time purchase. The current price can be checked on the retail listing.
• The app is free to download with basic features. According to its App Store listing, the full analysis and personalized programs require a subscription, with two tiers: sleep² PRO and the Intensive Program.
• No screen and no built-in "daily score". Unlike a ring or a band, there is no device here that tracks the whole day: it is a tool focused on the night.

Regarding the Intensive Program, the app listing describes it as a medical device approved in the EU for sleep disorders such as insomnia. That refers to that program, not to sleep stage measurement, and it does not extend the medical category to the rest of the app.

Practical limitations worth knowing

In the reviews on the App Store listing, which has an average rating of 3.6 from 37 ratings, opinions are split. One health researcher said that a night of data matched a lab study well and asked for more validation; another review complained that the app discarded an entire night after a brief Bluetooth signal loss. These are few ratings and cannot be generalized, but they point to things worth checking.

The developer itself acknowledges that recordings with more than 20% of the signal affected may be discarded, because they do not allow a reliable analysis. The comments on the listing also mention the lack of data export within the app (the developer places it among its priorities). And, for now, the listing we consulted is the iOS one; if you use Android, check compatibility before buying the sensor.

What you can learn from this and what you cannot

A more reliable stage record can help you see trends: how long it takes to fall asleep, how much wakefulness there is in the middle of the night, how the proportion of deep sleep and REM changes after routine changes such as eating dinner later or training in the evening. But three limits matter:

1. One night is not an experiment. Night-to-night variation is large, so comparing one night with a "supplement" and another without it does not allow any conclusion. If you test something, you need several nights in each condition and to change only one variable at a time. This is a methodological criterion, not a figure from any specific study.
2. Accuracy is not the same as usefulness. An 80% accuracy across four stages still means that a share of the periods is misclassified, especially in transitions between light and deep sleep. The exact minutes of "deep sleep" are an estimate.
3. Measuring better is not the same as resting better. None of the cited studies assesses whether using this app changes rest. The 2023 sleep training study had no control group and the authors admit that the subjective improvements cannot be confidently attributed to the program; the objective effect remained a trend in sleep latency (p = 0.086).

Who it makes sense for and who it does not

It makes sense if you have already been tracking your sleep for a while with a watch or a ring, you are interested in seeing the stages with a method that has published backing, and you are willing to pay for a subscription and put on a sensor every night. Also if you want to cross-check what another device tells you.

It does not if you want a device that measures everything (activity, stress, training) with no extra hassle; in that case, the guide to smartwatch, ring or band helps you choose a format. Nor if you want to know whether you have a disorder: that is what clinical polysomnography is for. And if what you are after is recording brain activity during meditation or concentration, that is a different category, such as the Muse 2.

How to try it sensibly

1. Choose the sensor based on what you will tolerate wearing every night, not on the highest figure.
2. Record for at least one or two weeks before changing any habit, to have a baseline.
3. Change only one thing at a time and record again for several nights.
4. Be wary of nights with an incomplete signal and check that the app flags them as such.
5. Use the stages as an orienting clue and always cross-check with how you feel during the day.