Getting enough sleep in the nights before learning protects what was learned: several nights of short sleep before encoding cause more forgetting of factual material weeks later than adequate sleep. Short sleep after learning also tends to lower later memory: across studies, restricting sleep after encoding impairs memory about as much as restricting it before (Crowley et al. 2024; mostly single nights; how long after learning memory was tested was not analysed), although the one multi-night study of factual prose retained over weeks found no clear increase in forgetting (Lo, Bennion & Chee 2016).
Status
weakened
Last change
2026-11-02
Serves
Q1 (How do humans actually learn, and what helps them learn best?)
Ages
gymnasium, vuxen-livslangt
Area
Sleep and the body
Bounds
Multi-night sleep opportunity before encoding of factual material, with retention measured days to weeks later; not school start-time policy itself (C14) and not retrieval practice (C0). Kill-condition test survived 2026-10-27 (Cousins, Wong & Chee 2019, Journal of Adolescent Health 65, 549–557, DOI 10.1016/j.jadohealth.2019.04.030; OA full text stored from lab mirror): adolescents 15–18, 5 nights × 5 h time-in-bed vs 9 h; learning after the fourth restricted night; more forgetting after short sleep (+26% at 30 min, d = .66; +34% at day 3, d = .69; +65% at day 42, d = 1.17); vigilance worse but not significantly correlated with the memory outcome. Weakened 2026-10-27 (Lo, Bennion & Chee 2016, Journal of Sleep Research 25, 664–672, DOI 10.1111/jsr.12424; OA full text stored): after encoding a prose passage, 7 nights × 5 h vs 9 h TIB; no main effect of group on free recall (F(1,43) = 1.13, P = .29); authors report no significant memory differences at any time point (initial, 1 week, 6 weeks) between SR and control for highlighted, non-highlighted, or total content; only the growth of the prioritization benefit from immediate to 1 week was significant in control (d = 0.99) not SR (d = 0.26). Diekelmann & Born 2010 (Nature Reviews Neuroscience 11, DOI 10.1038/nrn2762; abstract): sleep optimises consolidation (review; full text not yet read). Voderholzer et al. 2011 (Sleep Medicine 12, DOI 10.1016/j.sleep.2010.07.017; closed from the box; cited by Lo 2016 and Cousins 2019 as a post-learning consolidation null) remains a future lead, not applied. Spacing bound and further support (corrected 2026-10-28): Huang, Deshpande, Yeo, Lo, Chee & Gooley 2016 (SLEEP 39(9), 1681–1690, DOI 10.5665/sleep.6092; PMC full text): Singapore boarding-school adolescents 15–19 randomised to 5 h or 9 h time in bed for 7 nights while learning 40 GRE-type word pairs; words crammed in one session were recalled worse after short sleep (Tukey P < .001), but words spaced over four days were recalled as well (P = .80; spacing × sleep LR χ² = 8.40, P < .01), and the gap on crammed words remained after one review pass (P < .01; spaced P = .99). In an exploratory split by study day, crammed words learned after the last normal night were recalled as well (P > .99) while those learned after restricted nights were worse (P < .01, P < .001) — consistent with both halves of the claim. Limits: restriction came both before and after each study day, so it is not a clean pre-encoding-only test; the set split was exploratory; the authors say the spaced items may have been over-studied (ceiling) and no power analysis was done. See open question 17. Weakened again 2026-11-01 (Crowley, Alderman, Javadi & Tamminen 2024, Neuroscience & Biobehavioral Reviews 167, 105929, DOI 10.1016/j.neubiorev.2024.105929; CC BY full text, Kent repository): 125 effect sizes from 39 reports (N = 1,234), 3–6.5 h vs 7–11 h for 1–14 nights; overall g = 0.29 (CI 0.13 to 0.44); restriction after encoding g = 0.36 (CI 0.16 to 0.56; k = 72; 0.25 without outliers) vs before encoding g = 0.24 (CI 0.02 to 0.46; k = 43; 0.21 without outliers); timing not a significant moderator (p = .40); number of restriction nights not a moderator (p = .57); Lo et al. 2016's own post-encoding outcomes coded in the direction of impairment (g = 0.09 to 0.41, not significant in that study). The claim's 'after learning does not clearly increase forgetting' is withdrawn as a general statement; it is kept only as the result of the one multi-night factual-prose study. Limits of the meta: 60% single nights, only one effect size beyond 7 nights, nearly all non-declarative outcomes post-encoding, mostly single nights; the retention interval was not coded or analysed (Crowley §2.4, §2.7.5). In the combined restriction plus total-deprivation analysis, restriction before encoding gave g = 0.20 (CI 0.01–0.39) against 0.63 for total deprivation (§3.5); mean post hoc power 15.41% (range 7.61–25.19%); Egger asymmetry z = 3.90 with no bias in the selection model; not preregistered (§2.1, §3.2, §4.3). The first sentence's "weeks later" still rests on Cousins et al. 2019.
What would kill this
A randomised study finds that several nights of restricted sleep before encoding do not increase later forgetting of factual material compared with adequate sleep. Promoted from K35 on 2026-10-27 under the project's promotion rule when C20 reached the survival floor of 3; standing policy Jonatan approved 2026-10-14). Area: Sleep and the body (distinct from C14 later school start).
Tutor consequence
none yet
Attack record
3 attacks: 1 survived, 2 narrowed
2026-10-27 survived
Attack
A randomised multi-night sleep-restriction study around learning would meet C21's kill-condition if several nights of short sleep did not increase later forgetting compared with adequate sleep; Cousins, Wong & Chee 2019 found adolescents given 5 nights × 5 h time-in-bed (vs 9 h) forgot more factual material at 30 minutes (+26%, d = .66), day 3 (+34%, d = .69) and day 42 (+65%, d = 1.17).
Target
kill-condition
Source
Cousins, Wong & Chee 2019 | Journal of Adolescent Health 65(4), 549–557, DOI 10.1016/j.jadohealth.2019.04.030
Steelmans to
C21 dies if a randomised study finds that several nights of restricted sleep around learning do not increase later forgetting; this is the claim's own supporting RCT once the OA full text is read, and it directly manipulates multi-night TIB around classroom-like factual learning with weeks-later retention.
Why
Extra10 attack pass 2026-10-27 Europe/Stockholm (parent: never-targeted C21 first honest OA full-text attack(s); C20 already at floor). Full text read and stored (CC BY-NC-ND OA; sleepcognition-lab.org mirror of the Journal of Adolescent Health PDF). Parallel groups, N = 59 (SR 29 / control 30), ages 15–18; learning of arthropod facts after four restricted nights; surprise day-42 subset SR 14 / control 22. Kill fails: short sleep increased forgetting at every delay. Limits: boarding/protocol setting; Singapore; encoding after restriction (not only post-learning consolidation); vigilance impaired but not correlated with memory. Not a preprint. Not a restatement-only Bounds pad: ledger was empty and the OA full text had not previously been stored or used as a kill-condition trial. Claim sentence and Status still alive until the Lo weaken below. First survival (1 of 3).
Rewrite
none (bounds + practical rule only)
2026-10-27 narrowed
Attack
A randomised multi-night sleep-restriction study after material is already learned would undercut C21's broad "nights around learning" wording if short sleep after encoding did not increase overall later forgetting; Lo, Bennion & Chee 2016 found no main effect of 7 nights × 5 h vs 9 h TIB on free recall of a prose passage (F(1,43) = 1.13, P = .29) at 1 week or 6 weeks.
Target
claim-text
Source
Lo, Bennion & Chee 2016 | Journal of Sleep Research 25(6), 664–672, DOI 10.1111/jsr.12424
Steelmans to
C21 asserts that several nights of short sleep around learning cause more forgetting weeks later; this OA RCT restricts sleep for a week after encoding (consolidation window) and tests overall recall at 1 and 6 weeks — the strongest available OA challenge to treating pre-learning and post-learning short sleep as the same.
Why
Extra10 attack pass 2026-10-27 Europe/Stockholm (second applied attack after Cousins survival). Full text read and stored (CC BY OA; lab mirror). 56 adolescents 15–19; encoding then 7-night manipulation; free recall initial / 1 week / 6 weeks. No main effect of group; authors state no significant memory differences at any time point between SR and control for HL, nHL, or total content. Prioritization-benefit growth from immediate to 1 week significant only in control (d = 0.99 vs SR d = 0.26). Authors cite consistency with Voderholzer et al. 2011 (closed from the box — not applied). Kill withheld: Cousins 2019 still shows a large weeks-later forgetting cost when short sleep precedes encoding. Narrowed instead: keep the pre-encoding multi-night restriction cost; drop a clear overall-forgetting cost of multi-night short sleep after the material is already learned. Not a restatement of Cousins. Not a preprint. Status alive → weakened. Claim and Bounds rewritten same turn; brief and the practical rule updated.
Rewrite
claim-field+brief
2026-11-01 narrowed
Attack
A 2024 meta-analysis of 125 sleep-restriction effect sizes found that restricting sleep after encoding lowers later memory (g = 0.36; 0.25 without outliers) at least as much as restricting it before encoding (g = 0.24), with timing and number of restriction nights not significant moderators, so C21's statement that short sleep after learning "does not clearly increase overall forgetting" is not tenable as a general claim.
Target
claim-text
Source
Crowley, Alderman, Javadi & Tamminen 2024 | Neuroscience & Biobehavioral Reviews 167, 105929, DOI 10.1016/j.neubiorev.2024.105929 (CC BY full text read 2026-11-01, Kent Academic Repository; Crossref-checked)
Steelmans to
C21's second sentence generalises from one study (Lo, Bennion & Chee 2016, n = 45, not significant) to "the same multi-night short sleep after the material is already learned"; the meta-analysis pools 72 post-encoding effect sizes, finds a reliable impairment with no evidence of publication bias, finds no difference between before- and after-encoding restriction and no dose effect of nights, and even codes Lo et al.'s own outcomes in the direction of impairment — so the asymmetry the claim draws (before matters, after does not) is not supported.
Why
Nightly-attacks 2026-11-01 Europe/Stockholm. Full text: 39 reports, N = 1,234; overall g = 0.29 (0.13 to 0.44), 0.23 without outliers; no publication bias; Table 2: after encoding g = 0.36 (0.16 to 0.56; k = 72), 0.25 (0.09 to 0.42) without outliers; before encoding 0.24 (0.02 to 0.46; k = 43), 0.21 without outliers; timing Q = 0.71, p = .40; restriction nights β = −0.02, p = .57; declarative 0.30. Table 1 codes Lo et al. 2016 post-encoding prose outcomes as g = 0.41, 0.20, 0.09, 0.23. Kill withheld (and not applicable to the first sentence): 60% of effect sizes are single nights and only one is beyond 7 nights; nearly all non-declarative outcomes are post-encoding; most tests within days — so for multi-night restriction and factual material retained over weeks, Lo et al. 2016 remains the only direct test, and the claim keeps it as a stated result rather than a general rule. Not a duplicate (Lo 2016 entry weakened in the opposite direction; this attacks the sentence that entry added); not an attack on a withdrawn clause (the second sentence is live text). Not a preprint. Status stays weakened. Claim field and Bounds rewritten in the same turn; brief, practical rule (rule text and limits; headline unchanged — it speaks only of sleep before learning) and diagram core rewritten.
Rewrite
claim-field+brief
Corrections to our notes
These fix a number or a fact in our notes. They are not attacks and do not count toward the floor of 3 survivals.
— C21's limits do not yet record Huang et al. 2016, a randomised week of 5-hour vs 9-hour nights during vocabulary study in which crammed words were recalled worse after short sleep but words spaced over four days were not, and crammed words learned after the last normal night were recalled as well later.
— C21's claim text and limits say Crowley et al. 2024 covered "tests within days" / "mostly short delays"; the full text never codes or analyses the retention interval (a closer reading of the study).
Indirect, post-learning arm: sleeping right after learning word pairs (normal sleep, not restriction) gave no recall advantage at 24 h or 7 days (Exp 2a 7 days: slept-first 34.2% vs awake-first 37.1%; Exp 2b no sleep effect); delay groups not fully randomised.
Meta of 125 effect sizes: restriction before encoding g = 0.24 (supports first half); after encoding g = 0.36 (0.25 without outliers); timing not a moderator; 60% single nights. Weakened C21's post-encoding clause 2026-11-01.
OA Need-for-Sleep vocabulary RCT: 5 h vs 9 h TIB across a study week; massed recall hurt under SR, spaced protected (spacing×TIB interaction). Sleep after each study day — Bounds/spacing lead only; not applied as pre-encoding kill/survive.
Post-encoding 7 nights × 5h vs 9h: no main effect of group on prose free recall (F=1.13, P=.29); overall forgetting not raised at 1 or 6 weeks. Weakened C21 2026-10-27 extra10.
Cross-over RCT, 47 youth 10–14: 5 nights 7.5 h vs 10 h in bed before encoding lowered lure discrimination (F(1,46)=4.74, p=.035) but not general recognition (p=.38) 20 min later. Pictures, short delay, mild dose — indirect; not a kill-condition test.
Review state: first-pass draft, pending human review.