A new social-science paper delivers a surprisingly unsettling result: give dozens of expert teams the same data and the same ...