Albert Steppi

> This is more or less what the current infrastructure does. It has [bugs](https://github.com/scipy/scipy/issues/19524) that will be exacerbated by using `Generator` instead of `RandomState` (no cheap global instance to default...

@rgommers, do you think it's OK that I merged this before `array` vs `array_like` was settled? I saw your positive comment [here](https://github.com/scipy/scipy/pull/21041#issuecomment-2188583744) and thought it would be fine to merge...

> Changing the code to match `array` instead would be a larger change. It's very rare that we add a new submodule, so this is an opportunity to do it....

In https://github.com/scipy/scipy/pull/20834 for `nsum`, @mdhaber is updating the documentation from `array` to (appropriately types) `array_like` in both `nsum` and `differentiate`, which I think is an acceptable resolution.

I can take a closer look at the actual issue when I have time, but just want to voice that I'm against deprecating and removing`chdtr` and friends. I don't think...

I've been working on rewriting cdflib in C++, with the solver written in a regular structured style instead of the reverse communication style the original authors used to get around...

This was fixed incidentally in #22441 when we added a strict `NaN` check to `igamc`, which `chdtrc` calls https://github.com/scipy/scipy/blob/56c8b23935b0e09632bdbbfa6d105945db4f8279/scipy/special/xsf/cephes/igam.h#L328-L333 Do we need to add a test to confirm before closing?

I don't think it's a blocker to not have implementations for $I_{\nu+n}/I_{\nu}$ for general $n$. Since these two functions would be genuinely useful in `stats`, we could make them private...

> Thanks @steppi ! I opened this issue as a standalone function in `scipy.special` to draw more review attention :-) It's a good idea to put it private in `scipy.stats`...

I don’t have time to give detailed response today but 0) the consensus here is to never use numerical recipes as a reference. The licensing is onerous and It’s not...