And somebody must own the retained set, otherwise the carefully labelled images will outlast the parts they describe.

Who loads it for that comparison? The normal operator may seat it differently from the person who built the holder, even when both think they followed the same instruction.

Usual loader will do the comparison with the fixture owner. Quality has the retained bezels for physical examination. No new thresholds or lighting change today.

A repeat-placement check can establish image variation without settled defect labels; just don't call the moving feature harmless on that basis.

Keep that distinction in the capture sheet: which part and pose, what the image did, and the separately supplied surface decision.

Yes, Victor. My question was about repeatability, not deciding that any line which moves cannot be a scratch.

Quality confirmed a scratch on one retained bezel and accepted another with a bright streak. Repeated loading also showed two resting positions in the holder. Fixture owner is measuring which contacts each uses.

Did the operator recognise those two resting positions as different, or did both feel like normal seating? That affects the eventual loading instruction as well as the fixture review.

Both felt seated to them. It wasn't deliberate repositioning. We've recorded the contact observations with the frames; I can't yet say whether either pose is the intended one.

Then don't solve it by telling them to use the better-looking photograph's pose. Get the intended locating arrangement established and compare the image conditions from that basis.

And retain the scratch example in the subsequent comparison; a repeatable pose can still conceal the feature the inspection needs to detect.

The accepted bright-streak example matters as much, because it shows what normal finish the inspection must tolerate.

Can the fixture owner explain the intended loading in ordinary terms to the operator? I don't mean asking the operator to compensate for a contact problem with a more careful hand.

They reviewed it together. One support allows an alternative rest before the locating face engages. A fixture revision is proposed, but not made or tested. We have removed the guessed labels from the current reference set.

Keep the alternative-rest observation with the revision request. 'Improve repeatability' alone can lose the actual condition the designer is meant to address.

Attached the two contact views and their images. Thanks Mia for asking about repeat placement; we'd started a lighting argument before checking how the bezel sat. Capture repeatability and inspection performance remain unproven.

That is a useful stopping point for this set. Do not spend your separate validation examples on choosing the revised fixture and lighting.