I've fixed the internal linking issues through a two-stage process:

• When our original XML documents are processed to create the "clean" XML for public consumption, links to HTML pages and so on are turned into private URI scheme targets with a site: prefix and a corresponding prefixDef in the header.
• The HTML rendering now handles these links.

I've fixed a lot of individual links throughout the collection too in the process of doing that. The main things remaining are the issue of links between the references files, the naming of those files, and the question of whether we still need to offer the corpus.xml and corpus.txt versions for download.