I think that we constantly imagine moral universes that are not logical. Our legal systems represent such things, and they work hard to attain logical consistency through continual refinement, yet they evolve more internal contradictions all the time.
We generally imagine that we can state a system will all the constraints we would like to make us morally satisfied, and work out the conflicts later. But the potential conflicts are always essentially omitted from the concept of the system. In fact, the actions that render the morally compelling system logical and tractable arise ad hoc, and seldom resolve the actual logical problem causing the conflict until there are many, many instances of the same kind of compromise between moral principles.
I think that legal systems, and therefore the moralities they attempt to approximate, present a paraconsistent logic with only a local version of the law of non-contradiction in the same way that Intuitionism and other constructive mathematics present logics with only a local version of the law of the excluded middle (which applies, in those contexts, only when the options have been reduced to a finite number, or have been crowded in by proofs on all sides that reduce the problem to something essentially finite.)
So I would suggest that these two kinds of imagination converge on a single notion of sound human logic from opposite directions and that neither is a reasonable requirement to enforce on the other. We should not expect morality to be tractable, or mathematics to be humane. We need to live with systems that work despite not being complete in either way.
I would also propose that both of these layers of partial reasoning are forms of aesthetics. Mathematics is based upon its own feelings of consistency and clarity, sometimes referred to as essential elegance, and morality is based upon its own feelings of propriety and order, sometimes referred to as essential humanity.
So perhaps only the set of aesthetically appealing universes with a given set of interacting sources of value is really a good model. It can capture these two, and other human drives.