It's stated to be US-ASCII with 7-bit encoding, but it contains the UTF-8 character '—'. Fixing the encoding very slightly changes the average size of messages reported by mu:average, so adjust the new expected value.
