This continues my post that analyzed the Weng&Fan meta-analysis of the impacts of ChatGPT on "learning performance," "learning perception," and "critical thinking." Weng&Fan and the earlier Deng et al. study focused on "experimental studies" with the belief that these can generate robust evidence on the impacts of ChatGPT in education. Here I try to clarify some basic assumptions that must be made for such a project. If they are broken, the results are nonsense. That's why I have called these studies junk. Not a very polite thing to do, but as the claims of these papers spread virally, maybe a decent amount of antidote is needed in the circulation.