It's interesting that the thing to avoid is helping the wrong people. I might have naively thought that any nonnegative level of help provided to people other than the target demographic would be fine, and that outcomes could be compared primarily on the basis of the level of help provided to the target demographic.