It is an interesting question what restrictions there are on the
character number. 13.1.1 says:
The described character set portions must collectively describe
each character number in the described character set once
and only once.
Given this and given that the number is a "character number", I think
one could argue that the number in the character reference must be one
that was described (even if only as UNUSED) in the document character
set section of the SGML declaration.
I'm not convinced. The 13.1.1 text seems to be oriented towards preventing a
single character number from being described more than once than it is
oriented towards requiring every character number to be described at least
once. The text seems quite vague on this point. It is equally vauge re:
the extent of the DATACHAR class and how this relates to code extension.
Has this question been put to Charles or WG8?
Glenn