My own take as someone internal to that process is that it was a combination of 1 and 5.
I have no idea how candidates were screened. I do know that even before the “technical challenge” we had a large number of candidates completely faceplant on lowball questions asking what single line snippets of code did.
I can also say that I absolutely did not expect prod-ready results from the challenge. But I did expect things like not vomiting raw uuids on the screen instead of user readable values when displaying results. Or not having commits from overseas dev contractors which did all the actual work in your git log.