The XML has a self-closing <b/> tag, which is technically allowed by the schema. I believe it’s not valid HTML, though, which may be why some part of the Hugo pipeline turns it into a <b>. I guess we could (i) check if the schema can be written to disallow empty markup elements, and/or (ii) replace these with something valid in our MarkupText.as_html().
Originally posted by @mbollmann in #6407