Nice. Looks like this will address the use case of categories in
single-cardinality resource responses.

Yes, I think so.

Appendix B seems a little confused:

<snip>

The last sentence should say "where the cardinality is *not*
guaranteed to be one", I believe.

Right you are - I tinkered with the RFC2119 requirement levels to pass ID-Nits and must have messed it up while submitting it. Will be fixed in -01.

Sam