The short version
Sleep is the foundation everything else is built on, so we measure it carefully and tell you four things: how much you needed (a personal baseline, plus what you owe from past nights, plus a little extra after a hard day), how much you got, how good it was (a 0–100 score from duration, stage balance, and consistency), and what your night looked like (the awake / light / deep / REM breakdown). For devices that don't report sleep themselves, we even find the edges of your night from heart rate alone. This page explains every piece — the formulas, the constants, and the science behind the choices.
Why we’re publishing this
Sleep drives your Recovery score, shapes your next-day Strain target, and is the single biggest lever most people have over how they feel and perform. The numbers we put in front of you — “you need 7h 50m tonight,” “last night scored 82” — should be legible, not magic. Here is exactly how each one is produced.
1. How much sleep you need
“Eight hours for everyone” is a myth. Your need is personal, it changes with what you've been through, and it's the sum of three honest parts:
sleep need tonight = personal baseline + sleep debt + post-strain extension − naps taken
Personal baseline
Your floor — the amount your body needs on an ordinary day — is age-anchored:
| Age | Baseline need |
|---|---|
| Under 65 | 7h 30m (450 min) |
| 65+ | 7h 15m (435 min) |
These sit squarely inside the AASM / National Sleep Foundation adult recommendation of 7–9 hours.[1] We deliberately anchor to the lower-middle of that range as a baseline — debt and strain then push it up from there, rather than starting high and hoping you hit it.
Sleep debt — what you owe
Sleep loss accumulates, and you can't repay it all in one night. The science here is unambiguous: the performance and cognitive deficits from chronic sleep restriction build up in a dose-dependent way and do not fully reverse after a single night of recovery sleep.[2] So we track a running debt:
new debt = old debt + (minutes under baseline) − 0.8 × (minutes over baseline)
(kept between 0 and 150 minutes)Two choices worth calling out:
- Extra sleep pays debt back at 80% efficiency, not 100%. Banking an hour over your baseline doesn't erase an hour of accumulated deficit — recovery is real but incomplete. The 0.8 factor encodes that gradualism directly.
- Debt is capped at 2.5 hours. Beyond that, telling you to claw back ever-larger amounts in single nights is neither realistic nor healthy. We cap the debt and let you repay it gradually.
Post-strain extension — earning back a hard day
A demanding day genuinely increases how much sleep you need that night.[3] We compare today's Strain to your chronic (28-day) average and add extra sleep on a sliding scale — nothing for an easy day, up to +90 minutes after an exceptionally hard one:
| Today’s strain vs. your norm | Extra sleep |
|---|---|
| At or below 80% of normal | 0 |
| Around your normal | up to +15 min |
| ~30% above normal | up to +45 min |
| ~60% above normal | up to +90 min |
| Far above normal | +90 min (capped) |
Naps count
Any daytime sleep you've already logged is subtracted from tonight's target — you don't need to make up sleep you already got.
2. How good your sleep was — the Sleep Score
Each night gets a 0–100 score built from up to three components:
| Component | Weight (full data) | What it measures |
|---|---|---|
| Duration | 60% | Did you get the sleep you needed? |
| Stage balance | 30% | Was the composition healthy — enough deep and REM? |
| Consistency | 10% | Did you sleep on a regular schedule? |
When stage or consistency data is missing, the weights gracefully renormalize onto what's available (duration alone can still produce a score).
Duration score
We compare your effective sleep (time asleep, minus time awake) to last night's calculated need — so the target is your real, personalized number, not a flat eight hours. The score rises along a smooth curve that's gentle near your target and steep when you fall well short: hitting your need scores 100, and the penalty for missing it grows faster the further you drift below it.
Stage balance score
Healthy sleep isn't just long, it's well-composed. We score your deep, REM, light, and awake percentages against age-appropriate norms, because what's normal changes across a lifetime — slow-wave (deep) sleep declines markedly with age, and holding a 70-year-old to a 25-year-old's deep-sleep target would be both wrong and discouraging.[4] Our target bands (as a share of total sleep time) track the published lifespan trends:
| Age | Light | Deep | REM |
|---|---|---|---|
| 20–39 | ~55% | ~18% | ~22% |
| 40–59 | ~60% | ~12% | ~20% |
| 60+ | ~65% | ~7% | ~18% |
Deep and REM carry the most weight in the score (they're the restorative stages) and excessive time awake is penalized. These are population starting points — as you accumulate weeks of your own data, we shift the bands toward your personal history, so a stable individual pattern isn't flagged as a problem just because it differs from the average.
Consistency score
Going to bed and waking at regular times is one of the most robust, controllable drivers of sleep quality. We look at the spread of your bed and wake times over the last two weeks (once we have at least seven nights to judge from) and score tighter schedules higher — a routine that varies by two hours or more scores near zero on this component.
For users tracking a menstrual cycle, the luteal phase brings a small +2 adjustment, reflecting the lighter, more disrupted sleep that's normal in that phase.
3. What your night looked like — sleep stages
The awake / light / deep / REM timeline you see is produced by a machine-learning model trained on heart-rate patterns. As you cycle through sleep stages, your heart rate and its variability shift in characteristic ways, and the model learns to read those shifts into a stage for each segment of the night. We then stitch consecutive same-stage segments into the bands you see on your sleep graph.
A few honest notes on this:
- It's heart-rate-based, not a sleep lab. Clinical staging uses brain-wave (EEG) data we don't have. Wearable heart-rate staging is good at the big picture — roughly how much deep and REM you got — and less precise at the exact second each stage flips. That's a known and accepted limitation of the entire category, not just us.
- Stage percentages depend on how long you slept. REM is concentrated in the back half of the night, so a short night doesn't just have less sleep — it has disproportionately less REM. We account for this when judging your stage balance, rather than flagging a normal short-night pattern as a problem.[5]
- A night under four hours is treated as a nap, not a main sleep, and is handled separately.
4. Finding your sleep from heart rate alone
Some devices report sleep directly. Others — including some rings and chest straps — only give us a heart-rate stream. For those, we detect the boundaries of your night (when you fell asleep and when you woke) from heart rate alone, using a two-stage process:
- A coarse pass scans the entire 24-hour window in 5-minute steps with a model trained to separate sleep from wake, and identifies your main sleep period. We require a candidate period to be at least 45 minutes long, so a single quiet hour on the couch doesn't register as sleep.
- A fine pass then zooms in around the rough sleep-onset edge — looking at 2.5-minute resolution across a 90-minute window on each side — to pin down the moment you actually fell asleep more precisely.
Wake-up (offset) is read directly from the coarse pass. The model uses only heart-rate features (levels, trends, variability, time of day) and is careful not to over-extend the edges of your night.
What we’re honest about here
Detecting sleep onset from heart rate is genuinely hard when something keeps your heart rate elevated right up until you sleep — a late workout, a stressful evening — and in those cases onset can read later than it truly was. We know this is the weakest point in the boundary detector and it's an active area of improvement.
How sleep feeds the rest of the app
Your nightly Sleep Score is the largest single input to your Recovery score (30% of it), and your calculated sleep need is the target the next day's Recovery and Strain advice is built around. Sleep isn't a standalone vanity metric here — it's load-bearing.
What we’re honest about
What we’re honest about
- The baseline is age-based, not yet fully personalized. We use a robust age anchor for your floor and let debt and strain personalize from there. A future version may learn your individual baseline need from your own well-recovered nights.
- Stage detection is approximate by category. See section 3 — trust the trends (your typical deep / REM over weeks) more than any single night's exact minute count.
- Boundary onset can lag after a hard evening. See section 4.
- Consistency can over-penalize night owls. A very late but regular sleeper can show inflated schedule variance around midnight; we're refining how we handle that.
- Debt is a model, not a measurement. There's no universal clinical definition of “sleep debt,” so we chose an explicit, conservative one (gradual payback, hard cap) and documented it here rather than hiding it.
How this metric evolves
This is Sleep v1.0. We expect to personalize the baseline need, sharpen stage detection and onset timing, and keep tuning the score weights as we gather data and feedback. Material changes will be noted here.
References
- [1] Watson N.F., et al. Recommended Amount of Sleep for a Healthy Adult: A Joint Consensus Statement of the American Academy of Sleep Medicine and Sleep Research Society. Journal of Clinical Sleep Medicine, 11(6):591–592 (2015). doi:10.5664/jcsm.4758. Also published in Sleep, 38(6):843–844 (2015). Recommends adults regularly obtain 7 or more hours. National Sleep Foundation duration recommendations: Hirshkowitz M., et al. National Sleep Foundation's sleep time duration recommendations: methodology and results summary. Sleep Health, 1(1):40–43 (2015). doi:10.1016/j.sleh.2014.12.010 — adults (18–64) 7–9 hours. doi:10.5664/jcsm.4758
- [2] Van Dongen H.P.A., Maislin G., Mullington J.M., Dinges D.F. The cumulative cost of additional wakefulness: dose-response effects on neurobehavioral functions and sleep physiology from chronic sleep restriction and total sleep deprivation. Sleep, 26(2):117–126 (2003). doi:10.1093/sleep/26.2.117. · Belenky G., et al. Patterns of performance degradation and restoration during sleep restriction and subsequent recovery: a sleep dose-response study. Journal of Sleep Research, 12(1):1–12 (2003). doi:10.1046/j.1365-2869.2003.00337.x. Together these motivate gradual (80%-efficiency) debt repayment. doi:10.1093/sleep/26.2.117
- [3] Walsh N.P., Halson S.L., Sargent C., et al. Sleep and the athlete: narrative review and 2021 expert consensus recommendations. British Journal of Sports Medicine, 55(7):356–368 (2021). doi:10.1136/bjsports-2020-102025. The consensus supports individualized sleep extension and good sleep hygiene during demanding training, rather than a single fixed prescription. · Fullagar H.H.K., Skorski S., Duffield R., et al. Sleep and athletic performance: the effects of sleep loss on exercise performance, and physiological and cognitive responses to exercise. Sports Medicine, 45(2):161–186 (2015). doi:10.1007/s40279-014-0260-0. doi:10.1136/bjsports-2020-102025
- [4] Ohayon M.M., Carskadon M.A., Guilleminault C., Vitiello M.V. Meta-analysis of quantitative sleep parameters from childhood to old age in healthy individuals: developing normative sleep values across the human lifespan. Sleep, 27(7):1255–1273 (2004). doi:10.1093/sleep/27.7.1255. A meta-analysis of 65 studies (ages 5–102) showing slow-wave and REM sleep decline with age while lighter stages increase — the basis for our age-banded stage norms. For adult macro-architecture percentages see also Carskadon M.A., Dement W.C. Normal Human Sleep: An Overview, in Kryger M.H., Roth T., Dement W.C. (eds), Principles and Practice of Sleep Medicine, Elsevier. doi:10.1093/sleep/27.7.1255
- [5] REM episodes lengthen across the night (≈10 min early to ≈60 min late), so short nights under-sample REM; deep sleep is front-loaded into the first cycles. See Carskadon & Dement, Normal Human Sleep: An Overview (in Principles and Practice of Sleep Medicine, Elsevier), and the Ohayon et al. (2004) meta-analysis above.
This document describes a proprietary metric. The reasoning and constants are published in the interest of transparency; the underlying implementation is our own.