Sequential Testing

A family of statistical methods built specifically to let you check results as data arrives and stop early when the evidence is strong enough — without the false-positive blow-up that ordinary peeking causes. It spends your error budget carefully over time instead of all at once at the end.

The appeal is speed: a clearly winning (or clearly harmful) change can be confirmed in days instead of weeks, and you stop wasting traffic. The trade-off is that the “price” of being allowed to look continuously is slightly wider confidence intervals at any single moment. Common flavours include always-valid p-values and group-sequential designs with alpha-spending.