What you may be missing is how hard it is to actually create a good take home. Every take home test I’ve seen was rife with potential for the problem to explode in complexity. Even things that seem simple like names and dates have so many potential pitfalls that it can be impossible to tell as an applicant whether they intentionally laid a trap or not. Then there’s the incidentals, should I send them a docker image to increase the odds it’ll work on their machine? Oh, they’re going to want me to extend this in real time, should I include other nice to have services that this problem could dovetail into needing, like redis?

As far as I can tell, any company that isn’t willing to pay contracting rates to solve real problems on their stack is likely being disengenuous with their take homes and largely biasing against experienced devs who aren’t as likely to waste their time. And worse with a take home, is that they have no skin in the game. With an in person interview they lose at least as long as the inverview. With a take home they lose nothing except short email exchanges.