Thanks! http://about.validator.nu/#pitch seems to be using a RELAX NG schema instead of a DTD, which is fine, though it's not clear where the schema actually came from. The draft alludes to "Criteria that cannot be expressed by a DTD, but can still be checked by a machine", and that's a good sign. But I can't find any formalized list of those criteria, and that's really not.

And sure, there will always be overworked or ignorant authors who roll out slipshod work. But I at least want it to be possible to expect better, as it was not before we had the first DTD for HTML.