I get the impression that your models of decision theory implicitly assume that agents have logical omniscience. Logical uncertainty contradicts that, so any theory with both will end up confused and shuffling round an inconsistency.

I think to solve this you're going to have to explicitly model an agent with bounded computing power.