Sleep and next-day training capacity: what the evidence actually supports

Sleep loss measurably impairs next-day physical performance, but the effect is smaller and more selective than athlete folklore suggests, and the strongest claims in circulation rest on the weakest studies. The best current synthesis — a 2022 meta-analysis in Sports Medicine — finds real impairment across a range of physical tests, varying by task type and time of day. Maximal single efforts hold up comparatively well; sustained, repeated and skill-dependent work suffers more. Meanwhile the sleep-stage numbers on most wrists are the least trustworthy figures those devices produce. This is a review of where the evidence is solid, where it is thin, and where it simply runs out.

What does one bad night actually cost?

The most useful single reference is Craven and colleagues (2022), a systematic review and meta-analysis in Sports Medicine of acute sleep loss and physical performance. Its headline finding is that acute sleep loss impairs physical performance, with the magnitude depending on the outcome measured and on when in the day the test occurred — effects were generally more evident in afternoon and evening testing than in the morning.

That time-of-day pattern is worth pausing on. It implies the practical question is not only “did I sleep badly” but “when am I training, and how far into a day of accumulating wakefulness will I be.”

Underneath the meta-analysis sit older experiments that give the effect texture.

Reilly and Piercy (1994) partially restricted sleep in eight participants and tested weightlifting. Submaximal lifts degraded before maximal ones, and subjective effort rose before objective output fell. That dissociation — it feels harder before it is worse — recurs across this literature.

Skein and colleagues (2011) took 30 hours of total sleep deprivation into an intermittent-sprint protocol and found reduced sprint performance and pacing changes alongside altered muscle glycogen. Thirty hours awake is not a training scenario, but it establishes that the effect is not purely motivational.

Knowles and colleagues (2018) reviewed inadequate sleep and muscle strength specifically, and concluded that a single night of restriction has limited effect on maximal strength, while repeated restriction and multi-set work show clearer decrements. For a lifter, this is the most directly applicable finding available: one bad night is unlikely to move your top single much; a bad week will show up in your volume.

The broader narrative reviews — Fullagar and colleagues (2015) in Sports Medicine, and the 2021 expert consensus led by Walsh and colleagues in the British Journal of Sports Medicine — agree on the shape: cognitive and skill-dependent performance is affected earlier and more reliably than raw force production, and perceived exertion rises before measurable output falls.

Evidence grade: moderate. The meta-analysis is recent and well-conducted, but the underlying experiments are mostly small, mostly laboratory, and mostly acute. Chronic partial restriction — the actual pattern in most athletes’ lives — is far less studied than one dramatic night.

Does sleeping more make you better?

This is where the most-quoted study in the field also happens to be one of the weakest.

Mah and colleagues (2011) asked eleven collegiate basketball players to extend time in bed to ten hours nightly for five to seven weeks after a baseline period. They reported faster timed sprints, improved free-throw and three-point accuracy, and better reaction time and mood.

It is quoted everywhere, and it deserves its caveats stated every time. Eleven participants. No control group. Unblinded outcomes. A multi-week in-season block during which practice alone would be expected to improve shooting. The observed changes are consistent with a real sleep-extension effect, and equally consistent with training, familiarisation, and expectancy.

What it establishes: an intervention worth trying, at essentially no risk, with a plausible mechanism. What it does not establish: an effect size anyone should quote as a number.

The related concept of sleep banking has better experimental support. Rupp and colleagues (2009) extended participants’ sleep for a week before a restriction protocol and found the extension group better preserved alertness during restriction and recovered faster afterwards. For an athlete facing a known bad week — a tournament, a travel block — going in with sleep in hand is one of the few genuinely evidence-supported preparations available.

Evidence grade for sleep extension in athletes: low to moderate. Directionally supported, mechanistically plausible, quantitatively unestablished.

How much sleep does an athlete need?

There is no single correct number, and the more carefully the question is asked, the clearer that becomes.

Population guidance — the National Sleep Foundation’s consensus of 7 to 9 hours for adults — is a range, derived for general health, not a personal training target.

The individual-differences evidence is stronger than the population guidance. Van Dongen and colleagues (2003) established the dose-response relationship between cumulative wakefulness and neurobehavioural impairment, and showed that chronic moderate restriction produces deficits that accumulate while subjective sleepiness plateaus — people adapt to feeling fine long before they are fine. Their 2004 follow-up showed that vulnerability to sleep loss is trait-like: the same individuals are consistently more or less impaired by the same restriction, and the differences between people are far larger than the noise within a person.

The practical consequence is uncomfortable for anyone building an app. A fixed hours-per-night target is a population convenience. The honest object is each athlete’s own need and their own response, which requires their own data over time.

Athletes also do not sleep well by default. Roberts and colleagues (2019) meta-analysed the effects of training and competition on elite athletes’ sleep and found that intensified training and night competition disturb it — the periods where recovery matters most are the periods where sleep is most likely to be degraded. Vitale and colleagues (2019) review the practical hygiene measures with the most support.

Sleep and injury: association, not causation

The frequently cited finding here is Milewski and colleagues (2014), who surveyed adolescent athletes and reported that those sleeping fewer than eight hours on average had substantially higher odds of injury. A 2019 systematic review and meta-analysis in the same journal found the association held across the available adolescent studies.

Both are observational. Sleep-deprived adolescent athletes differ from well-slept ones in many other ways — training volume, school load, sport, socioeconomic circumstances — and the athletes who sleep least may simply be the ones doing the most of everything. The association is consistent and worth acting on cautiously. It is not evidence that adding sleep prevents injury, and no study has tested that directly.

Evidence grade: low for causation, moderate for association, and confined largely to adolescents.

Why your sleep stages are the weakest number on your wrist

This is the part most training advice ignores.

Chinoy and colleagues (2021) tested seven consumer sleep-tracking devices against laboratory polysomnography. Total sleep time performed reasonably. Sleep-stage classification did not — agreement for light, deep and REM stages was substantially poorer, with meaningful bias. Their 2022 follow-up under unrestricted home conditions reproduced the pattern in the setting people actually use these devices. The 2024 state-of-the-science paper from de Zambotti and colleagues sets out how such devices should be evaluated and what their outputs can support.

The implication is direct. If an app tells you to train lighter because your deep sleep was low, that recommendation is built on the least reliable figure the device produced. Sleep duration and timing from a modern wearable are usable. Stage percentages, at the individual-night level, are not a foundation for a training decision.

Where the evidence runs out

Three gaps are worth naming, because they bound what any product in this space can honestly claim.

Nobody has established the dose-response between a night of sleep and a training adjustment. There is no trial that says a five-hour night justifies removing one back-off set or capping RPE at seven. Every app that adjusts training from sleep — including Tuwa — is applying engineering judgment on top of a directional finding, and should say so.

Chronic partial restriction is under-studied relative to acute loss. The dramatic experiment is one sleepless night. The common reality is six hours a night for three weeks. The Van Dongen work is the best evidence here and it is neurobehavioural, not athletic.

Almost none of this is in the amateur multi-sport athlete. The subjects are collegiate teams, elite endurance athletes, or laboratory volunteers. The person who plays basketball twice a week, lifts three times, and has a job is not in this literature.

What this means inside Tuwa

Two things, and the second is more interesting than the first.

What ships today. Sleep contributes 25% of Tuwa’s recovery score, and that contribution is duration only, scored against a fixed 7.5-hour target: below 5 hours scores 10, the curve rises through 40 at 6 hours and 70 at 7.5, reaching 100 at 9 hours and above. Sleep is attributed to the day the athlete woke rather than the day the app happened to run, and the night is identified by clustering the raw HealthKit sleep samples so that an afternoon nap cannot be mistaken for last night.

That is deliberately cruder than this article’s evidence. It is duration against a population target, with no continuity, regularity or stage term — and given the Chinoy findings above, leaving stages out of a live score is a feature rather than an omission. But a fixed target contradicts the trait-like individual variation that Van Dongen established, and that gap is real.

What does not ship. A context-conditional sleep engine — stage-aware, with personalised need and explicit night profiles — exists in the codebase and runs shadow-only. It computes every night on real data, writes to local rows, and drives nothing an athlete sees. It is design in development, not a shipping feature, and no performance or accuracy claim is made for it here. It becomes a product claim on the day it is wired to the score, and not before.

To see how recovery signals are combined in the shipped app, read the readiness score.