A new (preliminary) draft of MADS is available at:
Similar to the previous draft, there is a companion version of MODS imported,
a slight change from the earlier version, and still compatible with MODS 3.0
Some changes in this draft from the November 23 draft:
1. The following is now explicit in the schema:
Each authority, related, or variant element has exactly one descriptor. Before, it could have more than one, but that was to accomodate name-title (so you included a name descriptor and a title descriptor) but aside from the name-title combination, there was no other case where it made sense to include more than one descriptor. What we've done is to add a nameTitle descriptor. (It combines name and title, and  also includes role, which is applicable only for a name-title, and role is therefore removed from the name definion.)
2. the <otherElements> wrapper is removed.
3.  <url> and <identifier> are moved into the descriptor, and are considered descriptor metadata  (removed from the miscellaneous elements, formerly wrapped by <otherElements>).
4. added "equivalent" as a <related> type.
Used for an authoritative term that is the equivalent of another in another authority system/thesaurus. Examples are a subject heading in a different thesaurus that is equivalent to the one being described or a
name used in another language-based authority file.
5. ID attribute added as suggested. Also a version attribute added.
6. Namespace error corrected.
7. xlink attribute added to <descriptors>.
8. <address> defined as "choice" instead of "sequence" so order is not prescribed.
9. madsCollection is now defined (despite the xmlspy bug).
10. <postition>, previously in <otherElements> (erroneously) is removed.
11. attributes, attribute groups, and other auxiliary structures are significantly simplified. This is due in part to the above mentioned change where there is only one descriptor per item, and in part to additional streamlining. Perviously there was allowed a set of attributes for each descriptor, and a global set for the item, but now as an item and descriptor are equivalent, one global set should do.  This may be oversimplified, if so we can change back.