Always fun to see your experiment called out in comments in a 3rd-party codebase: "However ATLAS complained that they sometimes have NaN's in <variable> and this interferes with some of their internal validation." What a bunch of whiners, ATLAS. :-)