Peeking, stopping early and testing twenty metrics until one looks good are the three classic ways to ship a "winner" that isn't. How to run a test you can defend.
No posts here yet.