>The very bimodal distribution is inconsistent with both options in your hypothesis (i.e. either the "terrible" or the "good" candidate).
it is completely consistent. When you fail at the very beginning you're a "terrible" and when you get offer you're a "good" one. The "terrible"/"good" isn't innate characteristic, it is in the eyes of beholder ... err ... interviewer. Most places pride themselves on taking only top 1|2%, and thus most candidates at any given place are "terrible" by definition, and thus the interview is built as the filter described by GP. Of course everybody hiring top 1% means that pretty much anybody would get to be a "good"
at some place.
it is completely consistent. When you fail at the very beginning you're a "terrible" and when you get offer you're a "good" one. The "terrible"/"good" isn't innate characteristic, it is in the eyes of beholder ... err ... interviewer. Most places pride themselves on taking only top 1|2%, and thus most candidates at any given place are "terrible" by definition, and thus the interview is built as the filter described by GP. Of course everybody hiring top 1% means that pretty much anybody would get to be a "good" at some place.