Proofing transcriptions is difficult. You can go a long way with computer models, but eventually you’ll get a feeling about a word not quite fitting and need to

Proofing transcriptions is difficult. You can go a long way with computer models, but eventually you’ll get a feeling about a word not quite fitting and need to check back to the source text. I guess you could run a high quality / more iterations OCR at this point, but more likely you’re relying on human proofing, and it doesn’t scale if you’re relying on humans. There’s a reason the Standard Ebooks project I contribute to has <250 books compared to Gutenberg’s 50k+: quality takes effort.