Pure Ruby statistics for deciding under uncertainty. Descriptive statistics, significance tests and distribution functions; Beta and Normal posteriors with exact P(best), expected loss and credible intervals; an A/B allocator with floors, fixed shares, an automatic win and a hold on it; estimators for values of several parts, pooled contrasts, hierarchical rates and a change test; and a generic Thompson-sampling bandit: a set of items with their views and yes/no or amount outcomes, ranked or picked by one rule from a key, given uniforms or an RNG, with the odds of what it draws: exact for a pick among up to 20 contenders (EXACT_PICK_MAX) and for orders and positions over up to 10 free items (EXACT_ORDER_MAX) with none barred from leading, otherwise a seeded Monte Carlo estimate with its standard error. Every random input is injected, so each answer can be reproduced.