Yes, images link to individually bagged faceplates. The scratch words were added by the person sorting the images, not by quality. We haven't got physical region-by-region findings alongside them.

Can quality examine those regions against the agreed appearance requirement before the image labels are used for anything further?

And keep uncertain separate from accepted. Removing a scratch guess doesn't award the part a pass.

Quality has now identified one actual surface mark for assessment and left the other streak regions unclassified. I've taken the guessed labels out of the tuning set, while retaining the original images and who labelled them.

Will the capture comparison include that identified mark once its product judgement is settled, so a lighting change is checked for visibility as well as a cleaner background?

Keep an accepted reference in that comparison too, Sarah. Otherwise every improvement only means fewer bright pixels.