Preregistration PB-1: population size and the mean-started descent in differential evolution (dose-response and subsampling of frozen populations)

Preregistration of experiment PB-1 for the manuscript "Unmasking the population mean: mean-started local descent for differential evolution on oscillatory big-valley landscapes" (in preparation). The plan, code, analysis scripts and pre-run test log were frozen by MD5 hashes before any run with the experiment's run seeds. Base algorithm: classic DE/rand/1/bin (F = 0.5, CR = 0.9), with and without the mean-started (1+1)-ES descent (KW) and a best-started control (KE). Targets: CEC 2017 F5, F7, F8 at D = 30 and 50, MaxFES = 10^4 D. Stage A (dose-response): population sizes 40, 2D, 4D, 10D, 20D; arms K0, KW, KE; 25 seeds (88401-88425). Preregistered verdict on the half-budget gain g = median log-error(KW) - median log-error(K0): MONOTONE if Kendall's tau between population level (40 < 2D < 4D < 10D) and g is at most -2/3 in at least 4 of 6 targets; otherwise WEAK if g(10D) < g(40) in at least 4 of 6; otherwise NONE. Stage B (subsampling): frozen populations of unmodified DE (NP = 10D; seeds 88301-88310; events at 10-50% of the budget). Descents of equal budget from the whole-population mean (W), the best individual (B) and the means of nested random subsamples of m = 5...160 individuals, with the same step size. Retained share rho(m) = (G(m) - G_B)/(G_W - G_B) of median improvements. Preregistered verdict: DYNAMICS if rho(40) >= 0.5; ESTIMATE if rho(40) <= 0.2; MIXED otherwise; UNDEFINED if the whole-population mean is not better than the best individual. Files: pb1_onkayit.tgz (MD5 f723f3c3bf0c50b434f0edddc8dc2fe3), plan kam-pb1-plani.md (MD5 9df6b626380bbd523e7c1ad888d37ffd, in Turkish), PB1_DONDURMA_MD5.txt (hashes of all frozen files, including binaries and third-party files that are not redistributed).