mean p = 0.1989;  fraction of prompts with p < 0.01: 0.276;  p < 0.1: 0.541
uniform k = 8: coverage 0.4937
oracle greedy: coverage 0.5632   samples per prompt: min 0, median 6, max 22, prompts with 0 samples 0.318
stop at first success, cap K = 13: average cost 7.794 samples/prompt, coverage 0.5557
budget  2: uniform 0.2936   oracle greedy 0.3764   stop-at-success (cap 2) 0.2936
budget  4: uniform 0.3957   oracle greedy 0.4696   stop-at-success (cap 5) 0.4282
budget 16: uniform 0.5804   oracle greedy 0.6508   stop-at-success (cap 34) 0.6612
budget 32: uniform 0.6552   oracle greedy 0.7256   stop-at-success (cap 90) 0.7451
