It's possible (under some assumptions) that this procedure, conditional on not sampling any power users, generates unbiased estimates of the CATE for non-power users.
It's possible (under some assumptions) that this procedure, conditional on not sampling any power users, generates unbiased estimates of the CATE for non-power users.