Your Apple Watch Just Got Better!
These notes were generated by AI and may contain inaccuracies.
Apple updated their sleep stage tracking algorithm across their entire lineup, from Apple Watch Ultra to Apple Watch SE, for better sleep stage tracking. This follows a recent video highlighting issues with deep sleep tracking on Apple Watch. Provides nuance and shows new results from Apple.
Days after uploading the deep sleep problem video, Apple added an appendix to their sleep stage tracking paper. Original printed version (2-3 weeks ago) ended without appendix; now online version on iPad includes it. Appendix states: in 2025, improvements made to sleep stage classification algorithm as part of iOS 26.0 and watchOS 26.0. Unclear exact rollout timing or if immediate with 26.0 worldwide.
Improvements Discussed
Discusses improvements, comparison to competition, and that it's still good despite ongoing deep sleep issues.
Rob, post-doctoral scientist specializing in biological data analysis.
Apple improved detection of quiet wake moments (very still before/after sleep but awake). Common issue across devices; example: sitting still in cinema misdetected as asleep. Less common now, but persisted with older Oura algorithm. Apple Watch uses only accelerometer sensor data (movement), not heart rate or beat-to-beat intervals, which sleep researchers say are important for good staging. Likely for battery life savings.
2025 updates permit full sleep stage classification: awake, REM, core, deep. Core is light sleep continuing after dismissing alarm. Previously, tracking stopped after alarm dismissal; now fixed.
Compares confusion matrices: reference on left, predicted (Apple Watch) on bottom (flipped from usual).
- Deep sleep: Old 62% agreement → new 68%. Confusion with core/light sleep: 38% → 32% (6% shift to deep).
- Awake: 70% → 79% (biggest improvement).
- REM: 78% → 82%.
- Core/light: 83% → 82% (stable).
Significant improvement. Scientific study used old algorithm; differences not huge but probably significant. Questions biological relevance and day-to-day matching. Good that Apple continues working on it.
Sad no heart rate data use, which could make it best tracker given Apple's resources. Deep sleep still an issue.
Nuance from prior video: Deep sleep issue exists, but Apple's numbers likely higher than independent validation with bad nights (e.g., alarms, sleep issues), which are harder to predict. Study archived; improving Whoop scoring with latest algorithm (retroactive possible for Whoop, not Apple).
Science takes time: by publication, Apple, Whoop, Fitbit/Google updated algorithms. Testing still valuable for latest devices; patterns match scientific studies (n=1, imperfect).
All wearables imperfect. Nuance: Apple Watch still one of best trackers even with old algorithm; overall agreement/Cohen's kappa similar to Oura Ring, though prefers Oura for consistency across stages. Oura more consistent per stage (better deep, but lower specificity for core/REM). Apple struggles more with deep sleep.
Day-to-day variation question: patterns may still be valid even if averages low.
Apple's improvements from internal testing with independent train/test sets. Happy Apple aware and improving; timing interesting (appendix ~1 week after video).
Asks viewers: noticed changes with watchOS 26 or post-release improvements?
Supports channel: YouTube membership, affiliates (Amazon, Eight Sleep Pod (favorite sleep device), Apple Watch, Oura Ring). Bookmark general Amazon link.
Other: Runna (2-week free trial, best running app), Strava Premium (20% off), YouTube membership (early access).
Considering full-time; thanks for watching/subscribing. Check prior Apple Watch sleep video or Eight Sleep Pod video.
Keep The Quantified Scientist in your library
Save the videos and channels worth coming back to, and find them again in one place.
