It's the implicit semantics of the tables. We start with modelling the world itself: "user with id ID is AGE years old and lives at ADDRESS", and end with modelling out particular instance of knowledge of the world: "user with id ID exists", "user with id ID is known to be AGE years old", "user with id ID known to live at ADDRESS"; and with the "closed world principle" you can explore more of the limits of the knowledge: "which users we know to exist and know their addresses but don't know their ages?". But query "what is the id of the oldest user?" is simply unanswerable unless you know all of their ages; the best you generally do is answer "what is the id of the oldest of the users that we know ages of?".
It's when we start confusing the map and the territory, that's when lot of miscalculations starts to happen.
I think you meant to write "open world principle", as every non state fact is implicitly false under the closed world assumption.
But yeah I agree, that databases should always be thought of a a model of the world (crisp), and not as the world itself (fuzzy).
But it's also noteworthy that this is a leaky abstraction, and that any kind of database that has to operate in the real world (in contrast to say an entity component game engine) will face this leakiness.
The only way I see to resolve this is to turn the database into something that remains crisp in the face of fuzziness.
E.g. by storing only observations or interpretations of the real world processes, since the observation is an "subjective" statement made from the database perspective, it holds true regardless of the actual "objective" condition.
It's just not easy to program and model in such a system.
But then again, the resulting model is factually a graph database and far far away from resembling something like SQL or Relational Calculus.
(Which might not be a bad thing though.
"You either die NOT NULL or live long enough to see an instance where you really wish that column was nullable.")