I live in Japan. I shoot film, write now and then, and spend some evenings with fragments of medieval manuscripts that sit in library catalogues as “unidentified”.
This site keeps all three. The finds are made with AI-assisted transcription and checked by hand; the people and libraries below are the ones who confirmed or acknowledged them.
Confirmed: A named specialist in the field has confirmed the identification. Acknowledged: The holding library has said the identification is new to its records. That is not peer review.
Fragments are chosen from digitised collections where the catalogue gives no identification or a doubtful one. A vision model transcribes each one blind, without the catalogue; the readings are matched against open text corpora; promising leads are checked by hand against the images and the reference tools, then sent to a specialist or the holding library.
Machine readings are corrected by eye but are not editions. A lead that finds no match is not thereby unknown, and the label on each finding says exactly how far it has been checked.