Recognizing these difficulties and biases, and watching the swaggering, overconfident, surface-level way coworkers around me talk about their interview candidates as if they had awesome powers of judgement, my approach has evolved into:

Pick an LSAT-style logic problem. First just give the candidate the problem, and see if they can work it out. No programming, just simple logical thinking skills. Then get into writing some code (or even pseudocode) that can solve some version of this problem.

There will be some conversation after that, maybe about how you'd scale the code, and then branching off into other things. But mainly I just see if a) they could solve the logic problem and b) they could translate that solution into some reasonable code. I try to apply this consistently across candidates.

I was unemployed once, and I knew that I know how to do this stuff. It was incredibly frustrating to me that the people interviewing me were holding me up to unreasonable and irrelevant standards. I keep that in mind and try to give people the benefit of the doubt.