Believe it or not, Python accepts arbitrary callables as metaclasses in class statements -- and those callables don't have to return instances of type. As long as the callable accepts the same arguments as type.__new__, it's an acceptable metaclass; the binding created by the class statement is set to whatever calling the metaclass returns:

>>> def returns_int(*args: object) -> int:
...     return 42
...     
>>> class A(metaclass=returns_int): ...
... 
>>> A
42

Currently we don't understand this in red-knot: we think that the class A statement constitutes a class definition, when it does not. Mypy "solves" this problem by emitting a diagnostic saying that a dynamic metaclass like this isn't supported. Pyright doesn't emit a diagnostic, and just gets it wrong.

I think mypy's "solution" here is acceptable (metaclasses that are not subclasses of type are really very rare). But it also might not be too much work for us to just add an accurate understanding of the runtime semantics here, and infer that the class A statement binds the A symbol to an instance of int rather than a newly created class.