I think it wasn’t even an OCR bug per se, it was a bug in the image compression algorithm implementation. “Yeah these squiggles on the paper look basically the same, guess we’ll save space here.”

OCR would have at least been mitigated by the fact that you could see if the text didn’t match the image. And since OCR isn’t perfect anyway, you could even anticipate that. But if the image is screwed up, well, what do you do then?