I can believe that quarks are ultimately responsible, but I'm not obligated to do so by a priori logical necessity.

I feel that someone should point out how difficult this discussion might be in light of the overwhelming empirical evidence for reductionism. Non-reductionist theories tend to get... reduced. In other words, reductionism's logical status is a fairly fine distinction in practice.

That said, I wonder if the claim can't be near-equivalently rephrased "it's impossible to imagine a non-reductionist scenario without populating it with your own arbitrary fictions". Your use of the term "conceivable" seems to mean (or include) something like "choose an arbitrary state space of possible worlds and an observation relation over that space". Clearly anything goes.

You're simply expanding your definition of "everything" to include arbitrary chunks of state space you bolted on, some of which are underdetermined by their interactions with every previous part of "everything". I don't have a fully fleshed-out logical theory of everything on hand, so I'll give you the benefit of the doubt that what you're saying isn't logically invalid. Either way, it's pointless. If there's no link between levels, there's no way to distinguish between states in the extended space except by some additional a priori process. Good luck acquiring or communicating evidence for such processes.