So if I’m understanding this correctly, they did a extremely high resolution CT scan, but it’s not possible to determine the letters from that. However by using other fragments of scroll as a ground truth, you can train a model to pick up on subtle differences in the fibres where there has been ink. That’s very cool!