Originally Posted by Manichean
I'm guessing you'll have to adjust the level 2 detection XPath to account for changes occuring during conversion. Have a look at the XHTML inside the ePub to see what the relevant headings are tagged like.
Nope, that's not the reason. In fact, if I move the level 2 XPath to level 1 and delete the input field for level 2, I get a single-level TOC containing the level 2 headings, so this is not the cause.