runs = 10000; false positives with no treatment effect: random allocation 4.4%, first-caught to treatment 12.9%