We're looking to build quality OCR into MuckRock so every document gets it, but it's surprisingly either a) technically difficult or b) financially expensive, choose one.