I  like the proposal  but I would like to strongly suggest:
1. Continue to call index sets "index sets", and call their prefixes
"index-set prefixes".
2. Call the new prefixes, for relations and relation modifiers, " relation
prefixes", and "modifier prefixes". I'm suggesting this so that we
completely disabuse ourself of the illusion that we're recreating the notion
of attribute vectors. An index prefix,  relation prefix,  and a modifier
prefix could all be the same string.

(Aside: what you're calling "modifier", cql currently calls "qualifier".
Which do we want to call these?)

(Minor editorial suggestion: could you make first and second rejected
solution 1.1 and 1.2 instead of 2 and 3?)

--Ray