Converted user Non-converting user p-value trace (−log₁₀ p) α significance plane
⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.

The Peeking Problem: A/B Test Significance

Synthetic users stream into a control bucket and a variant bucket at rates you choose, a live two-proportion z-test tracks the p-value after every batch, and a growing 3D trace shows exactly when the result crosses the significance threshold. Switch between "peek every batch" and a pre-committed fixed sample size to watch, over dozens of repeated experiments, why checking a dashboard every day and stopping the moment it turns green inflates the real false-positive rate far above the advertised α — the optional-stopping bug behind a lot of unreliable mobile A/B test results.