The simple explanation makes sense, but I'm realizing there's a subtle point here that I should been more clear on.
They might not be filtering out the bad at the expense of the good. But filtering out some of the good in the name of saving the money it would cost to develop / administer more general assessments.
That's the p(interview_capable) piece whereas the trade-off you mention is the conditional probability (also important!).
