We're doing a different kind of discrete optimization at $job and see much the same: there are other, often difficult to quantify preferences over and above wha

We're doing a different kind of discrete optimization at $job and see much the same: there are other, often difficult to quantify preferences over and above what we naively thought we were optimizing for.