Even more spectacular than the "illuminated letter" problem [which is bad enough][and which I would hope most transcribers would avoid nowadays by choosing NOT to include GIFs for illuminated letters and other trivial "printers art"] you also have texts when the transcriber has chosen to leave some text in GIF only mode, and/or other text in GIF mode AND OCR'ed mode, such that the MOBI and EPUB versions may have 0, 1 or 2 copies of a particular entire paragraph of text. And/or the HTML was written in a non-linear form in which case the MOBI and EPUB versions may have 0, 1, 2 or N copies of any particular passage in the text. And captions on images may be retained in the image, included in the HTML, and/or included in the alt-tag meaning that a particular user with a particular reading device may see or hear the image caption 0, 1, 2 or 3 times.