Suppose an evil demon rules over a world for the first of two epochs. In this first epoch, he tortures anyone who fails to internalize a stringent rule requiring them (unconditionally) to greet others by punching them in the guts. (To be clear, the demon doesn't mind if by some psychological fluke this disposition fails to manifest in action. He just wants to make sure that everyone has the wicked disposition.) In the second epoch, the demon leaves the people to their own devices. Then, at the end of the second epoch, the world will end. These facts are common knowledge. What should the people do?
The best outcome is presumably for everyone to (i) initially internalize the gut-punching rule, (ii) hopefully fail to act on it (whenever possible without undermining the disposition), and (iii) change their internalized rules once the demon goes away.
Rule-consequentialism clearly agrees with (i). It also implies, contra (ii), that agents ought to act according to this rule, i.e. they ought to punch each other in the guts, even though no good comes from this. But what does it imply about changing circumstances? Once in the second epoch, does the society get to rewrite the rules, so that what was previously obligatory is now forbidden? Does it make any difference whether the demon had also incentivized our internalizing a rule according to which we should never question or revise his rules? (Though initially beneficial, the changing circumstances would mean this rule, too, no longer has high expected value in the second epoch.)
Suppose that the people can predict that they will end up changing the rules in these (forbidden but) beneficial ways. Can their current moral code still condemn those future actions as "impermissible"? Or do contemporaneous codes trump? (Compare my old 'Reflecting on Relativism'.)