What would you want a PDF-to-EPUB converter to preserve?

One thing I learned while building Reflow is that converting the file isn't really the hard part. A PDF can look like a properly structured book to a human while containing very little of the structure an EPUB actually needs.

Reflow currently tries to detect things like chapters, headings, page breaks and images, while also letting you manually split, merge, rename and reorder sections when the automatic detection gets something wrong.

I'm curious what matters most to other people when converting a PDF to EPUB. Is it structure and navigation, images, formatting, tables, something else?