C11 Spacing and interleaving

survived narrowedone dot per attack, oldest first

1. #1 narrowed 15 October 2026 · Kremer, Gray-Lobe, de Laat & van der Haar 2026 (preprint) · target: kill-condition

   In a year-long randomised trial in 62 Nigerian grade-5 maths classrooms, interleaved practice (about half of each day's practice problems drawn from earlier lessons in the term) raised end-of-term posttests by 0.28 SD over blocked practice with the same number of problems, but its effect on a cumulative end-of-year test covering the same material was 0.03 SD (95 % CI −0.28 to 0.34), significantly smaller than the posttest effect, so blocked practice matched interleaving on that delayed closed-book test within realistic classroom time and C11's kill-condition is met for the year-long horizon.

   Strongest reading and verdict reasoning

   Steelmans to: C11 dies on "a meta-analysis or large classroom trial in which massed or blocked practice matches or beats spacing and interleaving on a delayed closed-book test, within realistic classroom time"; this is the first experimental evaluation of interleaving run over a full academic year, school-randomised, practitioner-delivered through scripted lessons in ordinary classrooms, with the problem count held equal; on the cumulative test (content taught on average about 145 days earlier) interleaving gave 0.03 SD, nearly ten times smaller than the 0.24 SD predicted from the posttest gains (surrogate analysis), and quantile estimates suggest harm at the top of the distribution (−0.33 SD at the 90th percentile).

   Why: Nightly-attacks 2026-10-15 Europe/Stockholm. Full text read (NBER PDF, revised February 2026; not stored). Design: 63 Bridge Nigeria schools, one excluded, 62 randomised (one grade-5 classroom each; 28 interleaved, 34 blocked; strata with one school dropped, one interleaved school had no eligible students); 687 incumbent students, mean age 9.81. Two 45-minute maths periods a day; same "my turn" worked examples in both arms; in interleaved classrooms the "your turn" problems mixed the topic of the day with problems from earlier lessons in the same term (roughly half each, mostly from the preceding three lessons or the current 3–5-lesson unit), so the program changes spacing as well as interleaving (the authors say effects cannot be attributed to interleaving alone). Tests: 30 open-ended items, graded by another teacher. Termly posttests (retention intervals on average 39, 38 and 29 days): +0.28 SD (Table 3, Panel B), similar across terms (largest 0.31 SD on the third). Cumulative test, five days after the program ended, items mirroring the three posttests, content introduced on average 145 days earlier: +0.03 SD (95 % CI −0.28 to 0.34); test of equal effects p = 0.02. Students one SD below the mean: posttests +0.54 SD; on the cumulative test the 20th-percentile effect was 0.49 SD but the 90th-percentile effect −0.33 SD (p < .05). Follow-up: 97 % for the posttest index; about 78 % in comparison schools for the cumulative test, 4 points lower in interleaved schools (ns). Kill withheld: (1) the same trial shows interleaving beating blocked practice on tests taken about a month after the material was taught, which is a delayed closed-book test within classroom time, so the claim survives at that horizon; (2) the cumulative estimate is imprecise and the authors call the study inconclusive on cumulative performance and say moderate positive or negative effects cannot be ruled out; (3) one setting (low-cost private schools with scripted lessons, frequent tests that themselves give spaced, mixed practice to both arms), one subject, one grade; (4) working paper. Context, as reported by the authors: the longest earlier interleaving program in maths ran four months in Florida and found 0.83 SD after one month (Rohrer et al. 2020); a two-day intervention gave 0.34 SD ten weeks later (Ziegler & Stern 2014). Not a restatement: C11's Bounds name Brunmair & Richter 2019 (blocking better for words; similarity of categories) and say the comparison in realistic classroom time had not been checked study by study; nothing on the claim limited the horizon. First ledger entry on C11. Claim, Bounds and Ages rewritten same turn; brief and the practical rule rewritten.

2. #2 narrowed 16 October 2026 · Ebersbach & Barzagar Nazari 2020 · target: claim-text

   In a preregistered experiment with 235 university students, practising a maths procedure (permutation problems) in two sessions 1 or 11 days apart gave no better scores one or five weeks later than the same practice in one session (spaced vs massed B = −0.02, p = .91), so C11's claim that spacing beats massed practice on delayed closed-book tests does not hold for this kind of procedural practice.

   Strongest reading and verdict reasoning

   Steelmans to: C11 states spacing beats massed practice on delayed tests in general; a preregistered, adequately sized between-subjects test of exactly that contrast, with two delays (1 and 5 weeks) and the same tasks that once showed a spacing benefit (Rohrer & Taylor 2006/2007), found no spacing effect and no lag effect.

   Why: Extra run 2026-10-16 (extra 4) Europe/Stockholm, ordered by Jonatan. Reached under the targeting rules after C9 (3e), C4 and C7 were fruitless. Full text read (open access, Europe PMC PMC7201105; not stored). Design: 3 (inter-study interval 0, 1, 11 days) × 2 (retention interval 1 or 5 weeks), all between-subjects, preregistered (OSF); 273 recruited, 235 analysed (38–40 per cell; mean age 24; paid); computer tutorial on permutations, two practice sets of four tasks, final test of procedural knowledge. Robust regression (1,000 bootstrap samples): spaced (1 or 11 days) vs massed B = −0.02 (SE 0.18, p = .909); 11 vs 1 day B = 0.26 (p = .103); RI × spacing interactions p = .078 and .633; longer delay lowered scores (B = −0.47, p = .001; 5 weeks M = 1.63, 1 week M = 2.53). Kill withheld: not a meta-analysis or classroom trial; university students working individually on a computer; one procedure; Rohrer & Taylor 2006 found spacing helped on the same kind of task after four weeks. Not a restatement: C11's existing bound is about durability over a school year (Kremer et al. 2026) and its Bounds mention item similarity for interleaving (Brunmair & Richter 2019); nothing limited the spacing half to declarative content. Not a preprint. Claim, Bounds and Ages rewritten same turn; brief and the practical rule Limits updated. Status unchanged (weakened).

3. #3 survived 17 October 2026 · Murray, Horner & Göbel 2025 · target: kill-condition

   A preregistered meta-analysis of spaced versus massed practice in mathematics would meet C11's kill-condition if, in the studies embedded in real courses with delayed exams, massed practice matched spacing.

   Strongest reading and verdict reasoning

   Steelmans to: C11 dies on "a meta-analysis... in which massed or blocked practice matches or beats spacing... on a delayed closed-book test, within realistic classroom time"; this meta-analysis analyses course-embedded studies separately, the closest test of "realistic classroom time", and the year-long trial already on C11 (Kremer et al. 2026) shows the classroom advantage can fade, so a null was plausible.

   Why: Nightly-attacks 2026-10-17 Europe/Stockholm. Full text read (accepted manuscript, White Rose Research Online; stored privately). Course-embedded studies: 17 studies, 26 effect sizes, g = 0.24 (SE 0.038; 95% CI 0.155 to 0.324); removing any single study left the lower limit above zero; trim-and-fill corrected g = 0.27 (0.215 to 0.326); time from last practice to exam two days to five weeks, five weeks most common. Isolated-learning studies g = 0.43 (10 studies, 27 effect sizes); all spacing studies g = 0.28 (27 studies, 53 effect sizes); the difference between the subsets was not significant (p = .095). Kill-condition failed. Limits: all course-embedded studies recruited in the USA, 65% of their effects come from universities, 46% from theses or a conference paper, and in 53% of effects learners could look back at their work during practice; course-embedded spacing also mixes topics, so spacing and interleaving are not separated, and studies that only compared blocked with interleaved practice were excluded; exam delays reach five weeks, so the year-end bound from Kremer et al. 2026 stands. Its retrieval-versus-restudy result in mathematics (7 studies, 32 effect sizes, g = 0.18, 95% CI crossing zero) bears on C0, not C11; noted as a lead. Not a preprint.

4. #4 survived 3 November 2026 · Ruitenburg, Goos, Lambert, Kirschner, Jarodzka & Camp 2026 · target: kill-condition

   In authentic primary and secondary classrooms (not labs, not university courses), spacing and interleaving may give no reliable advantage over massed and blocked practice on delayed tests within realistic classroom time.

   Strongest reading and verdict reasoning

   Steelmans to: The strongest version: earlier meta-analyses mixed lab and university studies, and Donoghue & Hattie 2021 found smaller or even negative effects in schools; a meta-analysis restricted to real school classrooms with active comparisons and quality criteria is exactly the test the kill line names.

   Why: Nightly-attacks run of 2026-11-03 Europe/Stockholm. Distributed practice g = 0.20 (95% CI 0.03–0.36; 15 studies, 56 effect sizes, 2,010 pupils), g = 0.39 against massed practice; interleaved versus blocked g = 0.40 (0.10–0.71; 10 studies, 57 effect sizes, 1,902 pupils); no publication bias by Egger. Massed/blocked did not match or beat spacing/interleaving on average. Boundaries found, not kills: spacing against less-distributed practice not significant; spacing significant only for gap/retention-interval ratios of 0.10–1; interleaving not significant with immediate feedback during practice (p = .75; sparse cells, explanation called conjecture by the authors). Not a restatement of the Murray et al. 2025 survive (US maths, mostly university courses): new population (school pupils aged 6–18, several subjects) and both strategies.