I've tried specialising this further with:

@overload
def getmask(a: np.ndarray[_ShapeT_co, _DTypeT_co]) -> np.ndarray[_ShapeT_co, dtype[bool_]] | bool_: ...
@overload
def getmask(a: ArrayLike) -> bool_: ...

The _DTypeT_co seems redundant; maybe removing it helps?
These overloads are also overlapping in an incompatible way now, because ArrayLike is broader than np.ndarray, but the bool_ it returns is narrower than the return type of the other overload.

ah right, thanks! now it passes (except one which PyRight detects accurates but MyPy doesn't, similar to in another issue)

except one which PyRight detects accurates but MyPy doesn't

yea... mypy can be a bit "special" in some situations 😅

needs a rebase

Thanks Marco