Exactly, there's no concept of confounding at all. You use "real world tests" (whatever the hell that is) when you have an actual specific setup to test. You

Exactly, there's no concept of confounding at all. You use "real world tests" (whatever the hell that is) when you have an actual specific setup to test. You use a small model experiment like this to test one specific thing like poll vs. epoll.