Work stats should be based on device_diff not work_difficulty since non-shares haven't been filtered out yet