Not quite sure about this: isn't it good enough to test via astropy? (I'm wondering a bit more generally why we would need a test of this every day, it is not like there are that many changes to pyerfa!)

Well, maybe this workflow is really satisfying two othogonal needs: regular testing and producing nightlies.
I agree that the latter doesn't really need to happen daily nightly, given the slow development pace: maybe we should instead push a new batch only on merge ?
Regular testing on the other hand I think is important to keep: sure, pyerfa doesn't move much, but numpy does, and when something affects astropy, it's always nice to immediately know whether this package is also affected or not. That said, maybe daily testing is a bit too much still. How about going weekly ?

This particular PR shouldn't be merged as is now: upstream changes have made it at least partially redundant. Let me draft this for now. I'll revisit soon.

Yes, weekly may be a good compromise.

For posterity, this is done in #157

Now that numpy has cp313 wheels, this is completely unnecessary. Closing now.