Sample size & duration, two-proportion significance, and a Sample Ratio Mismatch check — the three calculations every A/B test needs, with the formulas stated. Same math as the standards: nothing hidden, nothing sent anywhere.
100% client-side — nothing is sent anywhere · works offline
Two-proportion pooled-variance formula (Fleiss): n/arm = (z1−α/2·√(2p̄q̄) + z1−β·√(p₁q₁+p₂q₂))² / (p₂−p₁)². Two-sided. Peeking early invalidates the alpha — decide the duration first, then run it.
Pooled two-proportion z-test, two-sided p-value; the confidence interval on the difference uses the unpooled standard error. A p-value is not the probability B is better — it is the probability of a difference this large if there were no true difference.
Chi-square goodness-of-fit (1 df) against the intended split. Threshold p < 0.001 — an SRM check runs on every experiment, so the alarm bar must sit far below 0.05. A detected SRM means the assignment or logging is broken; the test's results are untrustworthy no matter how significant they look.