That already gives us an enormous number of points of comparison. It is, for example, why almost no one who advocates a "simulationistic" perspective directly uses the word "realism", because they have understood that that term is inherently vulnerable to a very obvious attack, one that is almost impossible to defend against: it's a fantasy, it has magic and dragons etc. (Of course, I find that many, many, many of the arguments that attempt to evade this are simply "realism with more steps", e.g. trying to use "verisimilitude", which lacks the "it has to be like Earth" element...only to then smuggle back in the "it has to be like Earth" element through ideas like naturalistic reasoning or "well the correct starting point is Earth unless told otherwise". In other words, they are literally just restating "realism", but trying to make it sound like it isn't just "realism" restated by splitting a problematic concept into two parts that seem milder in separate form.)
One cognitive model of fiction argues (from evidence) that people maintain an internal representation of the real world, an internal representation of the imagined world, and a meta-representational layer that among other things keeps track of which is which.
Given that picture
W is the set of facts fitting the real world model
W' is the set of fictional facts fitting the imagined world model
A candidate fact F can be counted "realistic" if it is true in W' or if not in W' nevertheless true in W
Seeing as this is a cognitive rather than scientific model, inaccuracies are tolerated according to norms (the "folksy common sense" a poster once used to describe this sort of realism)
That this type of truth-telling is covered by possible worlds theory helps to see that there is an additional quality - "accessibility" - that tracks whether
W and
W' are sufficiently similar to make this sort of comparison. One could imagine another possible world -
W'' - and that it could be the case that some
F' was true of
W' or if not in
W' nevertheless true of
W''. (Although that wouldn't fit the cognitive model, which maintains worlds that are "accessible" from one another.)
I think this addresses your objections - that there should be both dragons (a fact whose truth is established through its membership in
W') and let's say apples (a fact whose truth is established through its membership in
W.) And this isn't something uniquely describing imaginary worlds in games, it extends to fiction. Even dropping the cognitive model aspects of it, the possible worlds description still holds.
