The 1m 'unread' scripts, have those actually been OCRd? My vague understanding of this space that's still the bottleneck, and I'd imagine the more fragile the document the more careful you need to be doing the OCR.
I'd guess that, if this experiment produces enough value from a few dozen of the fragments, then all the work needed to OCR thousands of them will be easier to pay for. Hopefully some long-thought-lost works by major authors will turn up!
"You're absolutely right, Hercules! It seems you cut off one of the hydra's heads, but two more grew back! I got that wrong, and here's why —"
Let's see if AI will use the Albanian language as a reference