We Measure
the Measurers.
Consumer sleep trackers promise clinical insights from your wrist. We run them against polysomnography (PSG) baseline to see which algorithms are hallucinating REM cycles, and which are actually reading your body.
The 90% Accuracy Myth
Open any fitness brand's press release, and you will see the phrase "lab-grade accuracy." This is a marketing contortion. A consumer photoplethysmography (PPG) sensor cannot read your brainwaves. It reads heart rate, heart rate variability (HRV), and movement, then guesses what stage of sleep you are in.
When compared to an EEG polysomnography test (the clinical gold standard), most devices fail spectacularly at differentiating Light Sleep from REM, and frequently confuse quiet wakefulness with Deep Sleep.
We don't read press releases. We run trials. Every device featured on Sleep Tech Lab has been worn alongside a clinical EEG headset for a minimum of 14 nights to calculate its true Pearson correlation coefficient (r).
Notice the divergence around 3:00 AM. The wearable algorithm interprets a drop in heart rate as deep NREM sleep, while the EEG confirms the user was actually in REM.
The Accuracy Ledger
Agreement with PSG Baseline (Pearson correlation coefficient)
| Device | Total Sleep Time | REM Sleep | Deep Sleep (N3) | Form Factor | Our Verdict |
|---|---|---|---|---|---|
| Apple Watch Ultra 2 | r = 0.94 | r = 0.81 | r = 0.62 | Wrist | Review |
| Oura Ring Gen 3 | r = 0.91 | r = 0.74 | r = 0.76 | Finger | Review |
| Whoop 4.0 | r = 0.89 | r = 0.72 | r = 0.58 | Wrist/Bicep | Review |
| Withings Sleep Mat | r = 0.76 | r = 0.41 | r = 0.45 | Mattress | Review |
The Protocol
1. Baseline Setup
We establish ground truth using the Dreem 2 EEG headband, currently the highest-rated consumer-available surrogate for clinical polysomnography.
2. Concurrent Trial
The test device is worn on the non-dominant limb for exactly 14 consecutive nights, generating a minimum dataset of ~100 hours of sleep per device.
3. Epoch Analysis
Raw data is exported in 30-second epochs. We run an automated script to align the timestamps and calculate Cohen’s Kappa and Pearson coefficients.
Why Deep Sleep
Tracking Fails.
If your tracker tells you that you only got 12 minutes of deep sleep last night, you should probably ignore it. Here is the physiological reality of N3 sleep.
Read the whitepaperThere are four stages of sleep: N1 (lightest), N2 (light), N3 (deep, or slow-wave sleep), and REM.
When you are in N3 deep sleep, your brain waves slow down to massive, sweeping delta waves (0.5 to 2 Hz). This is easily readable by electrodes placed on your scalp (EEG).
Your wrist doesn't have a brain. A smartwatch relies on an optical sensor (PPG) that shines a green or infrared LED into your capillaries to measure blood volume changes. It is looking for secondary physiological markers: a drop in resting heart rate (RHR), increased heart rate variability (HRV), and a total lack of accelerometer movement.
The problem? The cardiovascular difference between N2 (light sleep) and N3 (deep sleep) is incredibly subtle. If you are a healthy adult, your heart rate might be exactly the same in both stages. Consequently, algorithms rely heavily on the timing of the night (assuming deep sleep happens in the first half) to guess. If your sleep architecture deviates from the norm, the wearable fails.
Current Benchmarks (Q3 2024)
Apple Watch Ultra 2
Apple continues to quietly dominate the accuracy charts. Without making any grandiose "recovery score" claims, their baseline sleep stage algorithm outperforms dedicated fitness bands by a margin of 12-15% in REM detection.
- ✓ Unmatched TST (Total Sleep Time) accuracy
- ✓ Highest correlation for REM vs NREM
- ✗ Form factor is bulky for side-sleepers
Oura Ring Gen 3
A ring on the finger provides a significantly better PPG signal than a watch on the wrist due to capillary density. Oura leverages this superior signal quality to deliver highly accurate sleep/wake onset detection.
- ✓ Zero wrist-fatigue or light bleed
- ✓ Excellent temperature trend data
- ✗ Struggles with REM/Wake differentiation
The Physics of Signal Acquisition
You cannot defy physics. The closer a sensor is to a major capillary bed, and the tighter it is affixed to the skin, the better the signal-to-noise ratio.
Nearables (under-mattress mats, radar hubs) rely on ballistocardiography—measuring the micro-movements your heart makes when pumping blood. They offer zero sleep friction, but severely struggle if you share a bed with a partner or a large pet.
Wearables (rings, watches, bicep bands) offer vastly superior heart rate and SpO2 data, but introduce sensory friction that can actively degrade your sleep quality if you are sensitive to pressure.
Read the comprehensive guideFrequent Misconceptions
Can a tracker tell if I have Sleep Apnea?
No. Not diagnostically. While devices like the Apple Watch Series 9/Ultra 2 and Withings Sleep mat can track SpO2 drops and breathing disturbances, they are not FDA-cleared to diagnose Obstructive Sleep Apnea (OSA). A high disturbance rate on a wearable is grounds for a doctor's visit, not a self-diagnosis.
My device says I got 0 minutes of REM. Is this accurate?
Highly unlikely. If you truly got zero REM sleep, you would be suffering severe cognitive impairment. Wearables frequently mistake still, quiet REM sleep for Light Sleep (N2) because heart rate variability stabilizes during both. Check our database to see your device's REM correlation coefficient.
Are "Recovery Scores" scientifically valid?
Partially. Proprietary scores (like Whoop's Recovery or Oura's Readiness) are heavily weighted heavily on Heart Rate Variability (HRV). While HRV is a proven metric for central nervous system fatigue, the arbitrary 0-100 score wrapping it is a gamified abstraction, not a medical measurement.
Receive Lab Notes.
We publish raw data drops when new devices launch. No marketing copy. Just Pearson correlations, PSG graphs, and raw analysis. Sent quarterly.
Zero spam. Pure data. Unsubscribe anytime.