We split testing into distinct stages primarily because: