Try paperless-ngx. It can do OCR and has search.

With 99+% of hashes matching?

Yes.

Maybe you can search for the track in soulssek. Get one that matches in size, rename it and try to recheck the torrent. Or if you are lucky you can get the file with the same name already and only recheck.

Maybe not only metadata. Could be baddumps, overdumps or alt roms.

I expect that big AI corps are scraping all content they can, so I think that all instances are scraped. The “good” part about Lemmy (not like R****t) is that the rest of the world can scrap the data too. So the power of our data is not only for big tech corps.