I set up experiments with a stated hypothesis, a threshold agreed before the result arrives, and enough traffic to mean something. The point is deciding in advance what would change your mind.
