> usually decide what's plausible
The issue was that every time we thought we had a decent idea of the physical constraints, without handwaving the entire thing as ludicrous amounts of magic architecture (to our sensibilities), someone eventually noticed that it didn't work out. Not a subtle problem, more like "uh actually if you built a 3d model of what we just described there wouldn't end up being any path between the two floors".
Okay most were a bit more subtle than that one. But that one did happen.