I once wrote a tool that helps with finding mistakes in OCR'd fixed width text, https://blog.qiqitori.com/2023/03/ocring-hex-dumps-or-other-...

Basically it just clusters same characters and asks the human to find the problems, which is easy when you're looking at a series of pictures like ssssss5sss.

The UI is kinda least-effort. Should ask a modern AI agent to make it look nice and intuitive, sometime maybe.