I cannot choose the FR10 scratch examples from these moving arrows

Two are here with capture numbers. The other pictures came from a supplier's presentation.

Keep the supplier pictures out of your confirmed set for now. You can ask for their physical reference, but you already have two covers you can investigate directly.

Were the arrows added during inspection or afterwards by whoever made the slides?

Are the two local covers accepted or rejected physically? I'd want that answered before comparing how bright the arrows make them look.

Ask whoever made the supplier slides which cover and inspection decision each picture represents. They may have a useful reference set behind the presentation, even though it is not supplied with it.

Will Archer, fine to ask, but don't delay the local comparison waiting for it. Two traceable covers are enough to discover whether the arrows are even pointing at the intended features.

Amy, slides afterwards. Leo Adams, quality has one accepted cover and one with a recorded edge scuff.

Does the scuff sit where any of the arrows point? That's the bit I'd like to hear before another round of camera suggestions.

Use the whole cover view for that match. A tight crop can lose the features the loader uses to tell one edge from another.

Noah, yes. And don't move the arrow to whichever line looks strongest after each reseat; the physical scuff stays on the same part feature even when its image changes.

David, that's probably what happened here, but Leo hasn't checked the match yet.

Fair point. I meant the next comparison, not a finding about these slides.

Compared them at the bench. Strong arrowed band is on the accepted cover; the edge scuff is elsewhere.

Remove those guessed defect labels and retain the identified full views with quality's decisions. Then investigate a view that actually shows the edge scuff without turning the accepted band into a reject.

Has the supplier answered the reference question? Their examples might still be useful for discussing lighting, but I would keep that separate from the newly identified local pair.