All very true!
With regard to Kwong’s “killer question:” when I was teaching people ideas related to this at Google, I referred to this as the “Asshole Genie Regression Test.” That is, imagine that you’ve acquired a magic lamp, and a genie who will make you score fantastically well on that metric… but that genie is a bit of a dick. In what ways could you score well on that metric and actually end up in a bad situation?
While this description is intentionally a joke, it has a deeper reason: the iron law of business is that you get what you reward, and you reward what you measure. If you set a certain metric and reward people (implicitly or explicitly) for reaching it, they will put all of their efforts towards that metric — and not necessarily towards whatever you actually wanted to use that metric as a proxy for optimizing. The result is, unfortunately, nearly indistinguishable from an Asshole Genie.