I put three FR5 trial images beside the corresponding brushed trim plates. The training labels follow the bright stripe in each image, but that stripe shifts when the plate is reseated. Nobody has signed off those locations as physical defects.
I have stopped threshold tuning. The inspection team can provide reference parts, but who should own the link between their physical decision, our image and the lesson material?
Who currently adds the labels to the images? I would begin with that person and the inspection decision-maker together, using one identified plate and all of its views, rather than asking either team to repair a whole folder in isolation.
The demonstrator added them for the lesson. Inspection had not seen the folder. They have now judged the retained plate acceptable and are selecting an identified reject with a visible physical feature.
Will the comparison keep the accepted plate too? A rejected reference is useful, but the accepted one explains why a moving stripe must not automatically acquire a scratch label.
Yes, both are in the plan. Thank you for pointing that out. The demonstrator will revise the lesson examples against inspection's decisions, while our imaging review checks how reliably those features remain visible after reseating.
Can the person who normally loads the plate take part in that reseating comparison? Otherwise the demonstrator may make the reference look repeatable with a placement habit nobody has actually taught.