> Coding questions ideally yield an answer that can compile and run and return correct output for at least some valid sets of input. That's about as objective as it gets.

Except for coding style (incl. naming convention), code organisation, test coverage, arch style, level of abstraction..plus a myriad of other things. I don't know anybody who treat the code in a coding question as a black box. It gets about as subjective as you can get when reviewing, with the (objectively) correct answer weighted the least.

Shit code that gets the right answer rates worse than code which ticks all the above subjective boxes but gets the answer wrong.