To use a famous, well-known example relevant to the HN user base: Software productivity can't be measured with any simple metrics - essentially every attempt at

To use a famous, well-known example relevant to the HN user base: Software productivity can't be measured with any simple metrics - essentially every attempt at coming up with software output metrics can be trivially gamed, yet that hasn't stopped people from trying mightily (I'm probably showing my age in recalling how much effort was put into "function points" from a couple decades ago). At the same time, nearly all of us can identify developers who are highly, highly productive, and others who are not. We may not have great metrics for software productivity, but I can guarantee that Fabrice Bellard is about 100 times more productive than I am. Metrics are inherently an oversimplification in attempt to get complex phenomena down to a strictly ordered value, but the real world doesn't work like that. I think your mistake is thinking that just because things can't be reduced so a small set of numbers would also mean they can't be observed, and that isn't true.