The text of an office document (note: Document) comes from the PDF (56 passages, no note) the converter (note: Converter) produced, never from the document's own format (16 passages, no note). A reader (note: Reader) still opens the original file, for its native properties (note: Attribute), but the words that enter the search index (note: Search index), the recognition (8 passages, no note) and the comparison (15 passages, no note) of twin resources (note: Twin resources) are read from the PDF, page by page, with one PDF library; a PDF given as a source (note: Source) goes through the same reading, and a transcript (note: Transcript), which no converter touches, gives its cues (note: Cue).
One path means one behaviour to test and one place to fix: the position (23 passages, no note) of every word is a page or a slide (note: Deck) of the PDF the reader will open in the document viewer, so a mention cites (4 passages, no note) what the reader sees, and a format is supported for text as soon as it is supported for conversion (note: Conversion). The price is accepted: without the converter, or when a conversion fails, an office document has no text and stays a download (30 passages, no note), and the count of such documents is what the fail-on policy of the build (note: Build) watches.