C21 Sleep before learning

survived narrowedone dot per attack, oldest first

1. #1 survived 27 October 2026 · Cousins, Wong & Chee 2019 · target: kill-condition

   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).

   Strongest reading and verdict reasoning

   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).

2. #2 narrowed 27 October 2026 · Lo, Bennion & Chee 2016 · target: claim-text

   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.

   Strongest reading and verdict reasoning

   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.

3. #4 narrowed 1 November 2026 · Crowley, Alderman, Javadi & Tamminen 2024 · target: claim-text

   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.

   Strongest reading and verdict reasoning

   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.