The table would have a simple block structure. See my message in this
thread December 20 or thereabouts. Legal variant types need not be
identical, but can be put in equivalence classes e.g. {geographic,
hierarchicalGeographic}. I believe it makes sense to view variant as an
equivalence relation: Two variants of a given authoritative form are
variants of each other (transitivity). A MADS record should still make
sense if a variant form were made authoritative and the authoritative
form made a variant (symmetry). If the authoritative form is,
conceptually, an identical "variant" of itself, well, there you are.

The children of <related> should remain uncontrolled. With "other" being
a valid value of the type attribute, anything could conceivably
cross-reference anything else.


>>> [log in to unmask] 01/03/05 5:26 PM >>>
> > So perhaps leaving this unvalidated is
> > best?
> I'd say no.  I think it opens up more problems than it solves.

What I meant was, we'd need to come up with a table of which
authority/related combinations are valid and enforce based on that.
like that would be a nightmare (though I haven't really thought it
and we'd probably get the table wrong anyway.

But if we could ascertain that authority and variants are always the
type then I'd favor validating that.