does it not actually convert it to an epub on the fly - like the calibre viewer ?
Ah - seems not & I see the issue:
I put a fresh copy of the book into calibre, so there was only the one format, & pressed T.
sure enough it opened. & I see that "chapter" is being used within actual chapters, as a class for each sentence, like this:
<p class="chapter">She climbed down....
That must be confusing the default xpath structure detection formulae - so its not a "bug" as such - it is a case where the default structure detection needs to be tweaked