it's also true that the default of D&D is normal physics, so some level of realism is present that follows normal physics.
Clearly it won't mirror reality, because it would be too cumbersome to figure it all out to that level of detail, but normal physics is the default.
I want to modify what you say here, to strengthen the apprehension implied in your second sentence, by proposing that the default of D&D is whatever
seeming of physics the group sustains in virtue of their
Knowledge of physics
Interest in narrating physics
Deliberate and potentially systematic exceptions
As others have suggested, their knowledge and interests may mean that what they narrate can be best described as a folksy common sense. That fits my suggested principle of "least difference", because what I am referring to is least difference from group norms of beliefs about the real world. It is obvious, but also crucial to observe, that fictional worlds lack physics. Nothing obeys physics in fiction beyond what authors know and choose to narrate.
As to what authors know and choose to relate, they may make intended and potentially systematic exceptions to their norms. Picture a group who normally believe in magic. For that group it could be that what others call magical is just what they'd ordinarily expect. Spells, for them, may be grounded in normal beliefs just as making a cup of tea is for me. For them my proposed "magic" descriptor does no particular work. I contend that nothing prevents them from using other descriptors demarking exceptions... from their normal beliefs about the world.
A related group is that which has familiarised with and adopted some set of world beliefs that is supposed to be imaginary. Such as for a genre or a shared fantasy world. They would be able to say what should normally happen in a given circumstances and reject that unfitting to their adopted genre or imagined world.
Group baselines are probably whichever they hold most easily in common; and may be blended with and overwritten by other sets. There can always be accompanying sets of exceptions: what falls into normal and what into exceptional can (and evidently does) vary per group.