I like this article, but given the number of misreads (small, in the example given) it might have been easier to diff a couple of OCR'd versions and just look up the diffs manually.