Sequential testing

Sequential testing is the practice of evaluating evidence continuously rather than waiting for a fixed sample and judging once. When the evidence is decisive early, you stop early; when it is inconclusive, you keep collecting. For monitoring, that is the natural shape of the problem — data never stops arriving, and you want to react as soon as, but no sooner than, the evidence warrants.

The catch is that naïve sequential testing reintroduces the peeking problem. Doing it correctly requires methods whose error guarantees survive optional stopping — which is precisely what anytime-valid tools provide.

ValidAnytime is sequential testing done right: e-processes accumulate evidence point by point, an alarm fires the moment it is justified, and the false-alarm guarantee holds at whatever moment you chose to stop.