p8 prereg (4 Nov 19:30): search-corrected step test. For each of the 21 p4 ships, scan split dates (every candidate day, 5-day margins), joint fit with only that ship split, z=step/analytic robust se; statistic = max|z| over the scan. Null: shuffle the dates of that ship's pair-days (partners kept), rescan, 100 perms. p = (1+#null>=obs)/101.
P1: >=5/13 drifters p<0.02 (I expect several to survive; Resistance, Trafalgar, Jason, Curlew surely).
P2: <=1/8 controls p<0.02.
P3: null max|z| median is >=3 (i.e. the p7 3-se bar was too low once you search).
