[marcosdumay]

Not just any non-crud indentifier. You'll need some very reliable ones. The more extreme the (un)likeness of something, the more reliable indicators you need for dealing with it.

For example, say that 90% of your potential emplyees are crud. You have a position with 100 candidates and 10 are fit for it. Now, you apply a (very well correlated) identifier, it reduces the candiadtes to 50, keeping 8. You still have too many candidates, so you apply another filter, reducing them to 25 in total, with 6 good ones... You can see where this is going, with some luck you'll have: 4/12, 3/6, 2/3, 1/1; you can expect quite often to have: 3/12, 2/6, 1/3, 0/1; with some bad luck you'll have: 3/12, 1/6, 0/3, and no good candidate will make it even to the last round of selections.