I think one reason why I got stuck multiple times is that it is easy to forget a definition, and that seems to translate to Any. I would guess mypy gives at least a warning for that, maybe even an error?
Could we make sure we notice undefined variables in the stubs when running tests?

EDIT: as of writing gh-24224 may have left _NBitInt and the way the aliases work in a (presumably OK), but not ideal state.

@BvB93 I think we want to keep the 2.0.0 milestone for this, but it's probably not blocking for the release branch creation and 2.0.0rc1, right?

Correct, I'll see if I can make some time next week to review the issue but this is by no means a blocker.

The numpy.random annotations were updated in gh-26089, which is already backported to 2.0.0.

@mtsokol @BvB93 did you mean to leave this open because ctypeslib (and maybe something else) still needs reviewing?

@rgommers For the ctypeslib reveal test where we have a separate if for win32 for np.long, I don't think there's much to do about it - in CPython when intc is the same size as long, c_int is an alias to c_long:

https://github.com/python/cpython/blob/92397d5ead38dde4154e70d00f24973bcf2a925a/Lib/ctypes/__init__.py#L183

From my side there's nothing else left here.

Let's close it for now then, if there is something remaining it's a bug now.