This is a genuinely interesting story. Huge amount of automated work, potentially saving hundreds of thousands of human hours. BUT: with an error rate of 10%, and with the precise reading of *every* word mattering in such an exercise, it is meaningless, and for manuscripts of authority worse than useless, without a human checking every single word.
BUT: there simply aren't people, in 2026, with the expertise, the time, and the funding to check these 32,000 manuscripts at this level.
Welcome to Digital Humanities Slop.