Your Sleep Tracker Knows Less Than It Pretends To

Somewhere in Chicago this morning, someone glanced at their phone before they'd even opened their eyes fully and learned that they got 1 hour and 47 minutes of REM sleep, 58 minutes of deep sleep, and a "sleep score" of 82. The number arrived with the crisp, unblinking confidence of a lab report. No error bars. No asterisk. Just a clean pie chart, as if a piece of aluminum and glass on your wrist had spent the night reading your brain waves like a paperback.

It hadn't. That's not a knock on the technology — modern wearables are a genuine engineering achievement, packing an accelerometer, an optical heart-rate sensor, and sometimes a thermometer into something the size of a large button. But "impressive sensor package" and "clinical-grade sleep stage measurement" are different claims, and the gap between them is exactly where your sleep tracker's confident little pie chart starts writing checks its hardware can't cash.

This is the story of what's actually happening inside that device while you sleep, what it can genuinely tell you, and where the science says to raise an eyebrow at the numbers before you rearrange your evening routine around them.

Consumer wearables have none of that. What they actually have is an accelerometer that detects movement and stillness, a photoplethysmography (PPG) sensor that shines light into the skin to estimate heart rate and derive heart rate variability from the pattern of blood flow, and in some newer models, a skin temperature sensor and a pulse oximeter estimating blood oxygen saturation. From that bundle of movement, heart rate, and HRV data, a proprietary algorithm — built by Apple, Fitbit/Google, or Oura, and generally not published in full detail — makes a probabilistic guess about which sleep stage you're likely in during each stretch of the night, usually broken into short chunks called epochs.

That inference is not baseless — there are genuine physiological correlations between sleep stage and heart rate variability (REM sleep tends to show more heart rate variability and irregular breathing patterns; deep sleep tends to show slower, steadier heart rate and minimal movement). The algorithms are exploiting real signal, not pulling numbers from thin air. But correlation-based inference from a different signal is categorically different from direct measurement, and that difference is precisely where accuracy degrades.

One recurring, unglamorous finding across these studies is that wearables tend to have trouble specifically with "wake after sleep onset" — the periods when you're actually lying still but awake in the middle of the night. Because the devices lean heavily on stillness as a proxy for sleep, a person lying motionless but awake, scrolling their thoughts at 3 a.m., often gets misclassified as asleep. This is a specificity problem: the device is better at correctly flagging real sleep than it is at correctly flagging real wakefulness, which quietly inflates total sleep time and sleep efficiency numbers in a systematic direction.

Epoch-by-epoch agreement — comparing each 30-second window of wearable output against the simultaneous PSG scoring for that same window — is the most rigorous way researchers evaluate these devices, and it's a tougher test than comparing total nightly summaries, because two devices can arrive at similar total REM minutes for the night while disagreeing about which specific minutes were REM. Studies using this stricter epoch-level comparison generally find that agreement, while better than chance and improving with newer device generations, still falls meaningfully short of the reliability clinicians expect from PSG-to-PSG scorer comparisons.

There's also a subtler statistical issue: manufacturers understandably promote their best validation numbers, often from studies they funded or co-authored, while independent, arm's-length replications are fewer and more scattered across different device models, firmware versions, and comparison methods, making it hard to build one clean, consensus accuracy figure that applies to "sleep trackers" as a category.