After some thought, I've decided on the following strategy for handling the gazetteer, and implemented some of it:

• We generate TEI files for each of the "letters" of the gazetteer alphabet, giving us a source file for each of those pages. That's working, along with generation of the JSON. These files require specific markup practices which the schema now allows for, and use two new private URI schemes.
• These files are put into the site/xml/original folder.
• They are validated there along with all the others.
• Validation will require some tweaking to the copied originals and these new ones, to do a couple of things: expand the XIncludes, and tweak the schema PIs so that they point to the correct location.
• From these files, we generate another file which contains a list of all the locations, and for each location, a list of the variant spellings, and for each variant, a list of links to the documents that use it. This will be used as the basis for providing the variant spelling collection at the head of each location file, along with appendix items which can be turned into popups showing links to all the documents.
• Gazetteer XHTML5 pages will be generated in the normal way from the "original" source files, once those have been converted to standalones. This will require addition of specific templates to match the tabular encoding and particular URI schemes (molagas: and molvariant:) which I've introduced as part of this construction.