Serhiy, thanks for responding.

I agree on this being "third-party" when looking at the original post I made, and I've reported the issue to the relevant party (https://github.com/pandas-dev/pandas/issues/44504). But really, the bigger issue here, in my opinion, is that the equality-hash invariance is poorly documented at best, and that this poor documentation is the likely "root cause" of the third-party issue I described.

If you read further up the thread, you can see that I have some (pretty concrete) proposals for how to improve on the current situation. I'd like the documentation maintainers to consider them, in whole or in part. If you feel it is more appropriate, I can split it out into a new issue, but I wouldn't mind having a conversation about it here.

The Python ecosystem will benefit as a whole if we keep an open mind as to how such (fundamental) issues can creep into commonly used, infrastructure-critical libraries, developed and maintained by excellent engineers.