That's not really actionable. Even if my skills aren't up to pytester's standards, we still need to recruit developers now. And for what it's worth I doubt I'll ever be able to explain most of the tasks our business is doing what you refer to as "properly".

I joined a team two weeks ago that built a fairly complex machine learning pipeline. One week in I was told I needed to help interview a candidate.

I lopped off a block of the code, made it run without the rest of the code base and explained what it was supposed to do. This wasn't that simple, but could still be explained in about a minute to somebody with familiarity in ML (assumed; that's what we're hiring for). I then challenged the candidate to spot any bugs and implement a feature.

This really wasn't that hard. I've done the same thing twice before in two completely different industries.

Why would that be impossible for you?

Perhaps I just don't understand what you're proposing. If it's general enough to be presentable without any domain specifics, then it seems indistinguishable from a general algorithm like reversing a string. I don't use tasks quite as simple as string reversal but not far from it.

As I read it the main observable difference between your exercise and the kind of exercise you're arguing against is that you are starting with some baseline code that the candidate is supposed to modify. That seems like a pretty good idea, but I haven't used it yet.