Why does the way you evaluate a service depends on when you evaluate it