SYNTH-07 · Guide
Sleep, time of day and caffeine
Protect the sleep opportunity before any recovery tool, after one short night train in the morning or move the key session a day, nap 20-90 min before 16:00, keep caffeine out of the evening, and place hard sessions where the athlete's day and chronotype allow; the hours and cut-offs are practice, the direction is evidence
The question it answers
"How much should I sleep?", "I slept badly — do I still do the session?", "should I nap?", "how late can I have coffee?", "is it better to train in the morning or the evening?", "how do I handle a flight across time zones or a dawn start?" What to do with the rest of a broken week is SYNTH-72; reading a low readiness score is SYNTH-30; caffeine as a supplement is SYNTH-67; the warm-up is SYNTH-66.
Default (what to prescribe first)
1
Sleep comes before every other recovery tool. A night of ≤6 h lowered performance by a mean 7.56% across every exercise category (REF-204, A-, 69 studies, n=959, 89% men, single nights). Athletes are at high risk of short, poor-quality, irregular and mistimed sleep, and sleep loss can impair strength, speed, cognition and recovery from injury (REF-205, B+, narrative review, no effect sizes).
2
How much. Not from a study in this base; the base's own practice: a regular sleep opportunity of 7-9 h a night, more in heavy blocks, with a fixed wake time (within about 30 min, weekends included).
An expert consensus calls a one-size-fits-all recommendation (for example 7-9 hours/night) unlikely ideal and recommends an individualised approach that considers the athlete's perceived sleep needs (REF-972, B+, narrative review and consensus, elite focus) — so the range is a starting point, not a target.
What the base does hold: ≤6 h is the level at which the single-night loss was measured (REF-204); 8 male hybrid athletes who rated their sleep "good" averaged ~6.6 h (REF-521, B, pilot) — so ask for hours, not for an opinion of sleep quality (inference).
3
After one short night: move a key endurance session a day; strength, speed or skill work can go to the morning rather than the evening (SYNTH-72; revised 2026-10-12 after the full-text check of REF-204).
The abstract says afternoon and evening tasks were consistently impaired and morning tasks largely unaffected; the full text's pooled table shows AM −5.42% vs PM −8.31%, and endurance tasks were impaired at either time (AM −6.50%, PM −3.56%; not formally compared — the authors read it as irrespective of time of day) (corrected 2026-10-12); performance fell ~0.4% for every hour awake before exercise (REF-204).
Which of the two options is better was not studied.
4
Nap when the night was short or the day holds two sessions (the trigger is the base's own practice; the review does not stratify by prior sleep): 20-90 min, between 13:00 and 16:00, with 30 min awake before training — the authors' recommendation in a review of 37 moderate-quality studies in which naps did not compromise night sleep (REF-608, B, no pooled effect).
5
Caffeine: 3-6 mg/kg about 60 min before, possibly effective from 2 mg/kg (REF-350, A), only after a training trial (SYNTH-67). None within 6 hours of bedtime, as a floor: a fixed 400 mg taken 0, 3 or 6 hours before habitual bedtime each disturbed sleep versus placebo, and no longer interval was tested (REF-973, B; not athletes, sample size not in the abstract, smaller doses untested). Not from a study; the base's own practice: 8 h where the day allows.
6
Time of day: put the session where it will reliably happen. Not from a study; the base's own practice: with a free choice, hard sessions in the afternoon. In 21 healthy adults (16 of whom did the exercise tests, at 9, 12, 15 and 18 h), performance peaked at 15-18 h, varying with the type of exercise (REF-325, B, correlational, no training at different times).
7
Warm-up length and script: [SYNTH-66]. This entry sets no warm-up minutes.
Decision (who → what)
| Situation | What to do | Source and how sure |
|---|---|---|
| One short night, key session planned | Endurance key session: move it a day, or lower the targets rather than cancel; strength, speed or skill work: a morning slot; eat normally (revised 2026-10-12) | REF-204 A-; SYNTH-72, SYNTH-30 (practice). After 30 h awake, 10 team-sport men sprinted slower with lower muscle glycogen (209 vs 274 mmol/kg dw) and unchanged RPE (REF-384, B — total deprivation, running; transfer) — RPE will not warn, and carbohydrate should not be cut (that entry's inference) |
| Several short nights in a row | Not from a study; the base's own practice: make that day's key session easy and reassess next day | SYNTH-72; REF-204 covers single nights only. In 9 endurance cyclists/triathletes, three nights with time in bed cut by 30% (4.7-4.9 h of sleep) slowed a time trial by the third day (60.4 vs 58.8 min) with RPE not different (REF-943, B, crossover) — so judge the session by output, not feel (that entry's inference); weeks of short sleep are untested |
| The alarm or the session | Do not buy a dawn session with an early alarm after a late night | Consistent harm came from total deprivation and from waking earlier than normal; going to bed later was not consistently negative (REF-204) |
| Poor night before a race | Race as planned. Inference: most starts are in the morning, where REF-204's abstract says tasks were largely spared (its full-text table: AM −5.42% pooled; endurance impaired at either time (corrected 2026-10-12)); transfer from laboratory tests to a race is itself inference. Extend sleep on the nights before | "Banking" sleep before a planned short night may help (REF-203, C+, narrative review) |
| Chronically short sleeper | Lengthen the time in bed for weeks, not one night | A ≥10-h in-bed goal for 5-7 weeks gave ~111 min more sleep a night and a faster timed sprint (16.2 → 15.5 s) in 11 collegiate basketball players (REF-3500, B, single-arm, no control; no endurance outcome — transfer) |
| Evening is the only training slot | Train then. A light session ending 2 h before the usual bedtime went with a night within a few minutes of a rest day; a moderate one (which includes an easy hour or more in zone 2) went with sleep about 15 min later and 15 min shorter; a high or maximal one (a heart-rate score above 214: e.g. about 55 min or more at 80-90% HRmax, a race, or a very long session) went with sleep 25-40 min later and 20-30 min shorter. The authors advise ending exercise 4 h or more before sleep where possible, otherwise choosing a light session in that window (REF-1193, B+, observational). Not from a study; the base's own practice: when a key session cannot end 4 h before bed, keep it and expect a later, shorter night (the cohort measured sleep, not training outcomes), and read the next morning's low HRV as the session's footprint (SYNTH-30) | A single evening session did not harm that night's sleep in a meta-analysis of 23 studies in healthy adults; sleep-onset latency, total sleep time and sleep efficiency might be impaired after vigorous exercise ending ≤ 1 h before bedtime (REF-974, A-; single sessions, not athletes). Across 16 RCTs of morning versus evening exercise, falling asleep did not differ, but evening exercise went with somewhat more waking in the night (overall SMD −0.51, very low certainty, possible small-study effects), clearest at moderate intensity (REF-1095, A-, mostly non-athletes, evening sessions mostly around 18:00). In a year of wearable data from 14,689 active adults, a maximal-strain session ending 2 h before habitual sleep onset went with sleep 36.0 min later and 22.2 min shorter, a night heart rate 3.9 beats/min higher and HRV 8.3 units lower than a light one; light and moderate sessions ending 4 h or more before went with sleep within a few minutes of rest days, a maximal one still about 12 min later and 10 min shorter at 4 h and not later or shorter at 6 h (REF-1193, B+, observational, not athletes, two authors employed by the device maker). In 8 male hybrid athletes a resistance session before bed went with a longer sleep onset (29 vs 10 min) and less wake after sleep onset (31 vs 48 min) than an endurance session (REF-521, B, pilot; session-to-bed time not given) |
| Late chronotype, morning race or session | Shift wake time earlier beforehand, bright light on waking, morning caffeine; warm-up by SYNTH-66 | Late types were impaired on every cognitive and physical measure at 08:00 versus early types; maximal voluntary contraction peaked at the 14:00 test in early and the 20:00 test in late types (REF-606, B, 56 healthy volunteers; peak times from its full text). The countermeasures are that entry's inference, not tested |
| Evening race or late key session | The athlete chooses: full dose and possibly a worse night (400 mg within 6 hours of bed disturbed sleep in non-athletes, REF-973; smaller doses untested), or a lower dose. Not from a study; the base's own practice: start at 2-3 mg/kg | REF-350: genetics and habitual intake may shape the sleep and anxiety cost; individual response varies |
| "Caffeine does nothing for me" or it causes jitters | Compare caffeine with none on two matched hard sessions (practice); drop it if there is no gain | In 101 competitive men, only CYP1A2 AA carriers were faster over 10 km on the bike (4.8% at 2 mg/kg, 6.8% at 4 mg/kg); AC saw no effect and CC were 13.7% slower at 4 mg/kg (REF-910, A-). The older phenotyping note is unverified (REF-446, C+) |
| Daily coffee drinker before a race | No withdrawal needed (inference and the base's own practice: no entry tests pre-race withdrawal) | In 40 trained male cyclists 6 mg/kg sped a ~30-min time trial equally in low, moderate and high habitual consumers (REF-909, B+). A new daily low-dose habit removed the acute benefit in the caffeine arm of a trial of 18 low-habitual users after 28 days (REF-006, B) — a different question |
| Long race (>~2 h) | Not from a study; the base's own practice: 2-3 mg/kg before plus 2-3 mg/kg at 90-120 min | SYNTH-67; the position stand has no dose-response for ultra-endurance (REF-249, A) |
| Long flight, few time zones | Travel fatigue only: rest, rehydrate, normal sleep. Do not manage it as jet lag | REF-605 A, consensus of 26 experts |
| Several time zones crossed | Budget ~1 day per zone eastward and ~0.5 day westward; shift sleep toward destination time before leaving; timed light, preserved sleep, strategic caffeine and naps | REF-605 (rule of thumb from its full text; eastward is harder; the panel says evidence on management is lacking). Not from that paper; the base's own practice: eastward — bright light in the destination morning; westward — in the evening; naps on arrival short and before 16:00 (REF-608) |
| Multi-day ultra | Do not cut sleep to gain time; plan the sleep stops: SYNTH-63 | In 23 Race Across France riders more sleep per 24 h went with a higher ranking (REF-796, B, correlational). In a 600-km race nappers were slower, but sleeping was self-selected (REF-854, B) |
| Wearable reports a bad night, athlete feels fine | Believe the athlete; use the device for bedtime, wake time and total sleep only | Six wearables agreed with polysomnography 86-89% on sleep versus wake and only 50-65% on sleep stages (REF-364, B, 53 non-athletes, one night). False "poor sleep" feedback lowered self-rated alertness (d = 0.79) in 63 people with insomnia while a vigilance test did not change (REF-398, A-) |
How to progress / periodise
Before an A-race: sleep extension is a weeks-long intervention (REF-3500: 5-7 weeks). Not from a study; the base's own practice: 30-60 min more time in bed through the taper.
Travel: the adaptation days of REF-605 go into the plan before the flights are booked; not from a study; the base's own practice: sessions in the first days after arrival are easy.
Rehearse caffeine at race dose in training or a B-race, never first in competition (SYNTH-67).
When NOT to use it / red flags
Caffeine is not a substitute for sleep (the base's own practice; the ISSN stand notes it may help performance under sleep deprivation in some individuals, REF-249), and a readiness or sleep score is never used to overrule an athlete who feels fine (SYNTH-30). Do not announce a bad sleep score before a key session: the feedback itself changes how the day feels (REF-398; insomnia patients, self-report).
Sleep tracking that raises anxiety is stopped (base practice). In a 523-person general-population sample 3-14% met orthosomnia criteria — owning a tracker is one of the criteria, so every case is a tracker user by definition (REF-401, B+, cross-sectional; not athletes).
Poor sleep together with falling performance, low mood, missed periods or low libido is an energy-availability screen, not a sleep-hygiene problem (SYNTH-54).
Sleep problems were reported by 88% of 187 female endurance athletes aged 40-60 and ranked among the symptoms that hurt training most (REF-731, B, self-report survey) — ask; persistent cases go to a physician.
In overnight ultras, hallucinations or disorientation are a stop-and-assess signal (the base's own practice; the study observed slowed responses and visual hallucinations: REF-665, B, 17 UTMB finishers).
Not from an entry in this base; standard practice: insomnia lasting weeks, loud snoring with daytime sleepiness, or sleepiness at the wheel go to a physician.
What the evidence does and does not show
Shows: a single short night costs performance, more later in the day in the pooled data, while endurance tasks were impaired at either time (REF-204) (corrected 2026-10-12); naps may help and do not harm night sleep when taken 13:00-16:00 (REF-608); a longer sleep window over weeks went with better sprint and skill scores in one uncontrolled study (REF-3500); performance peaks in the late afternoon on average and late chronotypes are worse in the morning (REF-325, REF-606); a single evening session did not harm that night's sleep (REF-974), while over a year of free-living nights in active adults later and harder sessions went with later, shorter sleep in a graded way, and light or moderate sessions ending 4 h or more before bed went with sleep within a few minutes of rest days (REF-1193); unlike a single evening session, 400 mg of caffeine up to 6 hours before bed did (REF-973); habitual caffeine intake did not change the acute benefit (REF-909) and genotype did (REF-910).
Does not show: a nightly sleep target for athletes; the cost of weeks of short sleep (three nights are in REF-943); an endurance outcome from weeks of sleep extension (three extended nights made a time trial faster by day 4 in 9 athletes, REF-943, B); that training at a chosen time of day produces better adaptation; a caffeine interval beyond 6 hours or a dose below 400 mg (REF-973); the sleep cost of habitual evening training in athletes (REF-974: single sessions; REF-1193: active adults, observational); any tested jet-lag protocol in athletes (REF-605 is consensus); women — the sleep-loss pool is 89% men and both caffeine trials are men only.
Evidence gaps
A measured sleep need for athletes: REF-972 is a consensus that rejects a single target and gives no hours.
Chronic partial sleep restriction over weeks and endurance training. No entry; three restricted or extended nights are covered by Roberts et al. 2019, REF-943.
Time of day of training and adaptation. No entry; the circadian physiology once listed here (core-temperature nadir and peak, chronotype shares, caffeine half-life) was textbook material without an entry and has been removed.
Alcohol and sleep. No entry.
How this entry was checked and changed
Why this level
(acute sleep loss on an A- meta-analysis of single nights; jet lag on an A consensus whose panel says evidence is lacking; naps on a B systematic review without pooling; time of day and chronotype on two B studies in healthy volunteers; sleep extension on one uncontrolled B study and a C+ review; caffeine dose on an A umbrella review, genotype on one A- trial in men; the 6-hour caffeine floor on one B placebo-controlled study in non-athletes; evening exercise on an A- meta-analysis of single sessions and a B+ cohort of 14,689 active adults over a year; sleep-need hours and the regular wake time are house practice)
Content correction 2026-09-27
Circadian core points (temperature nadir/peak, 15:00-19:00 power peak, chronotype shares) -> no entry in the base measures them; labelled the base's practice, and REF-204's measured time-of-day finding added.
"Drake 2013: 400 mg 6 h pre-sleep reduced sleep by ~1 h" -> Drake is not an entry; abstract-checked: significant disturbance at 0, 3 and 6 h, the ~1 h figure is not in the abstract and is dropped.
Caffeine cut-off "8 h" -> 6-8 h per REF-350 (REF-203: last intake 14:00 for 22:00 bed); half-life 5-6 h traced to REF-350, "range 3-8 h" dropped.
Rule 1 "12:00-18:00 larger effect size", "3-5% lower AM output", "sleep latency +10-20 min" -> unsourced; labelled practice or dropped. Added REF-204: after a short night, morning sessions are spared and evening ones suffer. Superseded 2026-10-12: full text AM −5.42% vs PM −8.31% pooled; endurance tasks were impaired at either time (AM −6.50%, PM −3.56%; not formally compared — the authors read it as irrespective of time of day).
Rule 3 "sleep extension (add 30-60 min pre-event, REF-203)" -> the measured intervention is REF-3500's 5-7 weeks at a ≥10 h in-bed goal (n=11, single-arm, C+); the 30-60 min pre-event version is labelled practice. "Alcohol disrupts REM" labelled not from the base.
Rule 5 "eastward harder" -> kept as general chronobiology but flagged: REF-205's entry says the opposite and looks wrong. "Avoid competing 24-48 h post-arrival" -> REF-205's "arrive 2-3 days early".
Rule 6 "second dose at hour 2+" -> REF-350's split dose at 90-120 min; breakfast CHO labelled practice.
Cross-reference REF-373 ("chronotype affects pacing") removed -> REF-373 is a deprecated duplicate of REF-010 and neither says anything about chronotype; REF-361 "morning measurement standardization" -> that protocol was removed from REF-361, now cited for rolling-trend reading and confounders.
Scope "Section 17 (sleep)" -> Section 12B, where the sleep entries live.
(sync 2026-09-27, after the full-base pass) REF-350 "half-life 5-6 h", "CYP1A2 AA ~40-50% / CC ~10-15%", "30-60 min pre", "split dose at 90-120 min", "not within 6-8 h of sleep" -> none in the corrected entry (ISSN: 3-6 mg/kg, ~60 min pre, from 2 mg/kg, genetic variation in response); timings restated, the rest labelled base practice or textbook.
REF-203 "half-life 5-7 h", "last caffeine 14:00 for 22:00 bed", "wake time ±30 min", "20-30 min naps 13:00-15:00" -> deleted from the entry (now C+); labelled base practice.
REF-446 genotype shares -> the entry (now C+, no verified source) says they are not quotable; removed.
REF-3500 "graded C+" -> now B (uncontrolled pre-post). Level parenthetical "A-level sleep, caffeine and monitoring reviews" -> REF-350/REF-361 A, REF-204 A-, REF-205 B+, REF-3500 B, REF-203/REF-446 C+; B+ kept for the framework.
Level parenthetical "over A-level sleep, caffeine and monitoring reviews" -> the A-range anchors are REF-350 and REF-204 only.
Resync 2026-09-28: REF-205 claims no longer in the corrected entry -> "heat sessions no later than ~16:00", "~1 day per time zone" and "arrive 2-3 days early" relabelled base practice / general chronobiology; the "REF-205 states the opposite direction" flag removed (the entry no longer gives a jet-lag direction). The earlier note in this section tying "arrive 2-3 days early" to REF-205 is superseded.
2026-10-01: rewritten in the decision-framework format and widened into the base's sleep framework; the default is now "sleep opportunity first (7-9 h, house practice), after one short night move a key endurance session a day (strength, speed or skill work can go to the morning), nap 20-90 min 13:00-16:00, caffeine 3-6 mg/kg with a 6-8 h pre-bed cut-off (practice), session time by what the athlete can do reliably".Contradictions resolved: early-morning "longer warm-up (25-30 min)" removed — warm-up length is SYNTH-66; pre-race breakfast "1-3 g/kg" removed — SYNTH-69; short-night rule aligned with SYNTH-72 and SYNTH-30; caffeine habituation and genotype now rest on REF-909 and REF-910 (as in SYNTH-67) instead of REF-446 (C+).Jet lag, naps and chronotype, formerly labelled "not from a study in this base", now cite REF-605, REF-608 and REF-606. Folded in: REF-605, REF-606, REF-608, REF-325, REF-205, REF-384, REF-521, REF-909, REF-910, REF-006, REF-249, REF-364, REF-398, REF-401, REF-796, REF-854, REF-665, REF-731.Removed as textbook without an entry: core-temperature nadir/peak, chronotype shares, caffeine half-life, "avoid HIT < 4 h before sleep", heat sessions before ~16:00, alcohol and REM, the two worked race-day schedules. Level B+ -> B; Type -> Decision framework.
Verification 2026-10-01: all 23 cited REF entries opened in KB.md.Numbers that match: 7.56 %, 69 studies/n=959/89 % men, ~0.4 %/h awake, 20-90 min/13:00-16:00/30 min, 3-6 and 2 mg/kg/60 min, 209 vs 274 mmol/kg dw, ≥10 h/5-7 wk/~111 min/16.2→15.5 s/n=11, 29 vs 10 and 31 vs 48 min, ~6.6 h, 4.8/6.8/13.7 % (n=101), n=40 at 6 mg/kg, 1 and 0.5 day per zone, n=23, 86-89 % and 50-65 % (n=53), d=0.79 (n=63), 523 and 3-14 %, 88 % of 187, 17 UTMB finishers.Levels/designs next to refIds match the head blocks (no LEVEL finding).
Verification 2026-10-01: main defect — REF-943 (Roberts 2019, three restricted/extended nights, endurance time trial, B, n=9) is in the snapshot and is cited by SYNTH-72, but SYNTH-07 says four times that the base has no such entry (lead, 'Several short nights' row, 'Does not show', Evidence gaps).The Level line and Cross-references should also gain REF-943 (not included in pairs: Cross-references are out of scope and the Level line edit is the owner's call). REF-943's authors advise '>8 h per night', relevant to Default 2 ('How much'), which currently says 'Not from a study in this base'.
Verification 2026-10-01: Level B is acceptable — the two core defaults (sleep first; morning after a short night) rest on an A- meta-analysis; 13 of 22 Default/Decision rows are practice or mixed and are labelled as such. No LEVEL-OVERALL.
Verification 2026-10-01: title — 'the hours and cut-offs are practice, the direction is evidence': the nap hours are the review authors' recommendation (REF-608), not practice, and the caffeine-evening direction rests on Drake 2013, which is not an entry. Wording imprecision only; no title change proposed.
Verification 2026-10-01: REF-392 (sleep extension, [UNVERIFIED]) exists in the base and is correctly NOT cited here.
Revision 2026-10-01: caffeine cut-off changed from 'none within 6-8 h (practice, Drake 2013 not an entry)' to '6 hours as an evidence floor (REF-973, B: 400 mg at 0, 3 or 6 hours each disturbed sleep; no longer interval tested), 8 h where the day allows as labelled practice'. The entry supports only the 6-hour minimum and says 6 hours was still not enough; the '6-8 h' range was not in it.
Revision 2026-10-01: evening-session row now carries REF-974 (Stutz 2018, A- meta-analysis, 23 studies, healthy adults, single sessions): evening exercise did not harm sleep; the authors' 'might be impaired' caution for vigorous exercise ending ≤ 1 h before bedtime became the row's action. 'Finish as early as the day allows' stays labelled practice.
Revision 2026-10-01: REF-972 (Walsh 2020 consensus, B+) added next to the 7-9 h practice line: the consensus calls a one-size-fits-all 7-9 hours/night target unlikely ideal — the house range is kept as practice and reworded as a starting point, not a target. Not folded: the consensus's statement that 1-3 nights of partial restriction 'remains unclear' sits beside REF-204's pooled single-night loss; left as is.
Revision 2026-10-01: three Evidence-gap candidates (Walsh, Drake, Stutz) removed and replaced by what remains untested; lead and Level parenthetical updated; level B unchanged.
Revision 2026-10-08: folded REF-1095 (Souabni 2026, A-, meta-analysis of 16 RCTs, n=611): no clear overall morning advantage; evening moderate-to-vigorous exercise went with more wake after sleep onset (SMD −0.51, very low certainty) — into the evening-slot row. Default, decision rows and level unchanged.
Revision 2026-10-12 (owner decision): the short-night rule now splits by session type — a key endurance session moves a day, strength/speed/skill work may take the morning — because REF-204's full text shows endurance tasks impaired at either time of day (AM −6.50%, PM −3.56%; the authors: irrespective of time of day), while the pooled data favour the morning for other tasks. Level unchanged.
Revision 2026-10-26: folded REF-1193 (Leota 2025, B+, cohort of 14,689 active adults, one year of WHOOP data): the evening-slot row now gives the night after a session ending 2 h before bed by strain (light about a rest day, moderate about 15 min later and shorter, high or maximal 25-40 min later and 20-30 min shorter) and the authors' advice to end 4 h or more before sleep or keep it light; keeping a key session that cannot end 4 h before bed is labelled practice.Lead, Level parenthetical, Shows, Does not show and Evidence gaps updated; level B unchanged.