I see wonderful symmetry here. There are TrendyCos (hot startup unicorns), BigCos (established companies like google or amazon or microsoft) and UnknownCos (not in the limelight so there is not much information about them). Likewise there are TrendyDevs (hotshots who produce one heavily github-starred framework after another), BigCoDevs (multiple years of experience at one of the BigCos, probably were responsible for…
There's no magic to it at all. Have every candidate work on programming problems related to the work you do at your company. Have every candidate work on the same problems, and let them do it from home. Build and iterate on a rubric for grading those challenges. It is amazing to me that almost nobody does this, but: almost nobody does this. They have programmers write code on a whiteboard, or on some whiteboard-codin…
The candidate shares their screen with you, so you can watch as they solve the problem in their own dev environment. You can understand how they approach problems (quick and dirty, slow and methodical, lots of rewriting, etc), and you can ask questions at the end. You get a good sense for how they work as an engineer even before having to bring them onsite.
This seems to resolve the time asymmetry of take-home tests as well -- the interviewer spends as much time watching as the candidate spends working.
The only downside I can see is that you have to design your problems to take about an hour instead of the 2-5 hours you could imagine for a take-home test. But, you can break them up into multiple rounds, and give additional exercises to the candidates who do well on the first one, for no more total time cost than a collection of onsite interviews.
For what it's worth, I've done this at my startup and hired a great developer, and got very positive feedback about the process from the candidates I didn't end up hiring.