Ah, but there's the rub. Candidates are trying to please the interview panel. If you don't provide guidance, the odds that they'll just use whatever they think the interviewers most prefer are just as good, if not better, than the odds that they'll actually use whatever makes them most comfortable.
You said yourself that since some candidates pick a language that you don't know well, you can't really tell if the failure of a large number of those candidates is reflective of a bad test or just a mismatched candidate pool. IMO, if you're going to stick to the "pick any language" thing, you should at least find out and ensure that any language the candidate picks will have a fair shot.
> it can even be a better interview because we get more insight on how the candidate communicates and their ability to walk someone else through their solution.
> A potential other bonus is less biases leak through from an interviewer on "that is a strange way to do that in language X."
I would say that if you're worried that interviewers will load in biases toward their preferred shortcuts etc in a specific language, that you should be equally worried that some good candidates are being excluded for choosing the "wrong" language in an any-language-goes test.
Above, you mentioned that there'd be a positive response if a candidate used "Rust, Scala, or a lisp dialect" -- these are all relatively trendy. What if the candidate used nim, Pony, or some other language that hasn't pulled in to the hype superstation yet? What if the candidate used a language that's not-so-trendy anymore, like Visual Basic, Cobol, or bash? What if the candidate used a programming language of their own design, and brought a copy of the compiler with them on a flash drive?
I'm asking because I've seen this in practice. Candidates for a devops position who chose to use bash to implement the very simple take-home task they were given were laughed off by several other members of the interview committee, despite being potentially high-value senior people -- they were at least senior enough that they're more comfortable performing sysadmin-style tasks in a shell, rather than using a massive CM framework or an awkward amalgamation of Python scripts running os.spawn.
It feels like this type of thing happens a lot, in the same sense that very often, "unlimited PTO" just means "guess whatever amount of PTO is acceptable around here and hope you get it right".