Suppose there's a method for calculating a fake choice token from a true choice token. This means that applying it to a fake choice must also produce corresponding fake choice tokens, otherwise a fake choice could be identified in this way. And by induction, any number of applications of the function must also reach a matching choice.
(This isn't a complete proof, but chasing up the remaining loose ends makes me think such a system is impossible, or at very least extremely complicated, and almost certainly impractical for non-mathematicians.)
I vote for candidate 0. The voting machine tells me the random number 1 (out of 0, 1). When the voting results are published, I can check that next to my name it has 0^1 = 1. When someone asks me what my random number is, I can either answer 0, to pretend I voted for 0(random number)^1(public list) = 1, or 1, to say that I voted for 1(random number)^1(public list) = 0.
Or any variation of it.