I agree that the negative's are a very difficult case, but in my experience is that most people don't cover the easier case of candidates they can evaluate. Further, evaluation techniques themselves are pretty error prone. But without collecting data and experimenting with it, how do we improve the problem?