After working on diagnostics fixes, I see that there are some major changes we need to consider for consistency's sake in the XML:

• The references.xml and botanical.xml files respectively are converted to noms_propres.html and terms_med.html in the output. This is a bit confusing, and it means that links to those specific web pages can't be encoded as normal links to the source XML documents, as they should be. I propose renaming the XML files to match the HTML output, and globally changing all the links throughout the corpus.
• DONE: There are many cases where we want to link to HTML files that are built for the site, but which don't have XML source files (such as toc_gravure.html). These links in the HTML are of course pointed at nothing. I propose that we adopt a prefixDef of site:toc_gravure.html for such links, and dereference it as mariage.uvic.ca/toc_gravure.html. This also applies to schema and ODD links.
• FIXED: There are a couple of cases where the references/noms_propres file links to the botanical, and vice versa. Because we don't expect these links, those elements are not imported into the back matter of the files when the XML is expanded, but the links are converted to local links. In the website context, this doesn't cause a problem; when the target is not there, the JS just gets it by AJAX. But the XML documents are not strictly valid. How to handle this? Ideally we should import that stuff.
• DONE: The normalized texts remove all the forme works, but in cases such as Le Bon Mariage, these contain page numbers with @id attributes which are the targets of links in TOCs etc. Make sure these targets are converted into the marginal page numbers we use instead.