Pieces retained, but only two images have definite part links. The rest are a tray's worth of guesses.

Then the unlinked pictures are not defect references. Useful evidence of your imaging problem, possibly. Do not recruit the nearest faceplate to give them a respectable identity.

Can you repeat the two linked pieces with the presentation and light settings written down?

Luca, I'd include the seating check in that. A nominal robot pose doesn't tell us whether the faceplate actually sat against the intended contacts. I'm not claiming that's Ella's problem, but it would make repeat pictures harder to compare

Yes, and don't change the holder and lighting together unless you label that as a combined change. Otherwise the next person will not know what produced the improvement, or even which arrangement they are supposed to reproduce.

Quality accepted one linked faceplate and found a physical score on the other. Both still show moving bands.

Is the score's region visible in every view, or merely somewhere in your best photograph?

Amy's question is the difficult one. Removing bright bands might make pictures prettier while hiding the actual score. I'd keep the physically accepted faceplate in the comparison too, so you can see both mistakes

The score disappears in one view. Holder seats repeatably in the checked static setup. Imaging still fails that example.

That sounds more useful than the old scratch labels. Is the imaging supplier seeing both linked examples?

Ella, with the location marked on the full view as well as any close crop. I would ask them to explain what each proposed view is meant to reveal, not just send back one attractive image of the scored part.

They have both. Proposed two-view comparison now, not a chosen threshold. Each view must stay linked to its faceplate.

Keep the pairing intact in the test results. Two images from two convenient faceplates would answer a different question rather impressively.

Did the paired comparison show the score without rejecting the accepted piece?