Following your advice yesterday, I have been experimenting with the various stages of HTML output from the convert-ebook debug option. I have tried 2 different PDFs and found something strange with both of them.
The Parsed, Processed and Structure HTML versions all had far too much italic and bold when viewed in my browser (Firefox).
I finally tracked the problem down to a few strange HTML tags, namely <b/> and/or <i/> which appeared at the Parsed stage and still remained at the Structure-stage. They were not present in the Input-stage HTML.
Both these tags caused problems with the text following. If viewed in the browser all the remaining text following an <i/> was italic. Similarly, all the text following a <b/> was bold. So by the time there had been one of each, the remaining text to the end of file was bold and italic.
The good news is that when I manually deleted the strange tags, all the text became correct again in the browser, i.e. no intended bold and italic were lost, so I was able to carry on experimenting.
Are these strange tags meant to be there?