Comment by MaxBarraclough

In a corporate setting, the solution would presumably be to get permission from further up the chain of command than the individuals being experimented upon. I think that would resolve the ethical problem, as no individual or organisation/project is then being harmed, although there is still an element of deception.

I don't know enough about the kernel's process to comment on whether the same approach could be taken there.

Alternatively, if the time window is broad enough, perhaps you could be almost totally open with everyone, withholding only the identity of the submitter. For a sufficiently wide time window, Be on your toes for malicious or buggy commits doesn't change the behaviour of the reviewers, as that's part of their role anyway.