“The scanned images look correct at first glance, even though numbers may actually be incorrect.”
The compression swapped patches of pixels that looked alike, so a 6 became an 8 and the page still looked clean. The bug sat in Xerox firmware for eight years across hundreds of thousands of machines. OCR was switched off, so nobody could blame a text engine. Every archive scanned on those machines is now suspect, and nobody can say which pages are wrong.