Seriously, I rhink you'rre on the right track.

Interesting to note:

Your list has significant commonality with the standard DP stylesheet.

Many of the elements you enumerate are notionally and/or notationally identifed during the DP process, but are overtly discarded on the way from OCR to PG. The conscious intention, I suppose, is to give PG what it looks like, not what it is.

Esstentially what we start with is a single big long string of characters produced by OCR software. Even at that point many of your artifacts are identified or inferrable. Along the way we discard and add artifact knowledge strictly to accomplish narrowly defined local tasks. Finally, we get rid of almost everything except what makes it look nice.

We don't 3ven have explicitly identified chapter titles.

I've often seen "class="poetry" for stuff that had nothing to do with poetry.