Wrong, the system must support a *subset* of ISO 10646 to be
conforming. Numeric character references that exceed the system
character set code point range can be mapped in an entirely
application specific manner, and the system will still be conforming.
ISO 10646 support is primarily an abstract idea. If the application
see 〹 and simply uses the lower 8 bits as the code, this is
fine because we can say in the abstract that the system maps the
numeric character reference to the ISO 10646 character, and then maps
that to a system representation consisting of the lower 8 bits. The
mapping to ISO 10646 will, in reality, never take place on most
systems.