I'm making an OCR website focused on outputting ascii text that follows the layout of the original, so that it doesn't need to understand or interpret zones in the source: it just resembles the source. This makes proofing easier and should also improve feeding documents to LLMs.