We requested atoms in latin1 and then handled them as utf-8, erroring out on some valid atoms and performing pointless validation on others.