I much prefer the failing-out approach over the rejection-letter approach. With the latter, you exclude those who would excel but look bad on paper, in favor of those who won't but look good on paper.