Following the issue of the missing NFLD newspaper documents, we determined that the problem was bad linking in the documents themselves, which DH fixed; I then had to rebuild to schema to get things to validate. I also got tired of the tedious process whereby XML documents are validated one by one, something that was necessary because jing suffered from a stack overflow due (I think) to the complexity of the folder structure within which the XML files were found. I discovered that if I just copied the XML files to a temp directory in a flat layout, jing would validate them just fine, and I could also give it a bit more memory anyway by setting my ANT_OPTS like this in .bashrc:
I think the -Xss solves the stack overflow. This means that I can now validate the XML in a few seconds. I then started looking at the Schematron, which has never been run as part of the build. Borrowing from other more recent project builds, I'm now generating a static .sch file as part of the ODD build process, and then compiling that immediately to create an XSLT file, following the model of DVPP. The validation process then runs that against the document collection, stores the results in a temp directory, and a second process parses those results to generate errors and fail the build if necessary. Found and fixed several errors in this process.
So now the NFLD documents are included in the site, and our build process is WAY faster.