> > I also wonder why not use xml:id
> > instead of MADS's own ID.
> It's not MAD's own ID, it's xs:id.
I had noticed that the type of MADS' @ID was xs:ID, and it is the reason 
why I suggested the use of @xml:id.

> record when it is encapsulated
> within a wrapper record,
It seems that this problem could be better solved with xs:keyref and 
xs:key (these are a totally different things than xsl:key) which offers a 
flexibility unavailable with xs:IDREF and xs:ID together. In "ID/IDREF 
Versus xs:key/xs:keyref" pp. 149-150 in the book "XML Schema", Eric van 
der Vlist explains that quite well.

I mean that the usefulness of the type xs:ID is mainly both in the past 
and in the future. At the time when @xml:id was not yet available, ID's 
could only be expressed with the type xs:ID on a local element. In XML 
Schema 1.1, several attributes (with different names) will probably be 
allowed to have the type xs:ID even when these attributes are placed on 
the same element in an instance document, and therefore, xml:id will not 
be enough as an attribute name for attributes of type xs:ID. But XML 
Schema 1.1 is not (yet) a W3C recommendation.