>>and maybe no internet connection.

Even the top notch devs I know can't do anything meaningful without an internet connection. Unless of course you know how to invent several layers of libraries, instantly in hours without bugs. Know the documentation of the libraries memorized to the last full stop.

In my exposure at least I know 0 devs who can do this. And I'd bet even the interviewer wouldn't be able to write any meaningful without a network connection. This isn't 1960s. By very definition almost anything today requires an internet connection.

I love this suggestion, and thank you for elaborating! FWIW, the interviews that I've talked about taking as a candidate actually used this approach. And, when I give interviews, I use this approach too.

The only thing I'm really defending here is using a 2nd stage screening process that only takes 30 minutes, before I go to the 3rd stage that takes a half day or full day. @BitL seems to be fighting the screening process, which is a necessary practicality and isn't going away.

The first stage is reading resume and glancing briefly at any side projects, which is ~5 minutes. The first stage is invisible, and filters out more people than any other stage. If anything unfair is happening, it's happening here, and there's little a candidate can do about it.

My thoughts on your outline:

A - getting rid of the whiteboard is of course optional. In my case, having graded exercises doesn't mean I stop using the whiteboard. I only provide a whiteboard for people who want one, and I don't ask people to whiteboard things better done in an IDE.

B - Yes! I do like having exercises that range from super easy to not solvable. It helps to know where people get stuck.

C - Yes! That's very helpful for language specific jobs. Personally, I also like exercises that are language agnostic, to let the candidate pick the language they're most comfortable with. I've loved doing coding quizzes in Python during interviews, and would prefer that over, say, C++.

What I liked was using shared Jupyter Notebook during a video call. You avoid whiteboard coding, you can see what the algorithm does immediately and fix any issues and both interviewer and candidate can have more interesting conversation, telling you if you are a good personality fit.