The build was broken because the wonderful NVU validator was finding errors in the HTML5 output that originated in bad attribute values in the original XML (@xml:lang and @target values mainly). I've now tracked down all of those and fixed them, and also caught some other problems with head elements inside lists. The build is now working again, and I'm now writing XSLT to generate a really stripped-down version of the website, by transforming the existing output, designed for rapid crawling by the archive-it crawler.