Bezel images taken with two different camera crops

Quality has it now; they found a faint line on the physical edge, but have not decided whether it exceeds the cosmetic limit.

Leave its label undecided until they do. You can still compare the images without asking the uncertain sample to teach the system what a reject is.

Farah, both crops came from one source frame; I've saved the regions, and quality is comparing the bezel with the agreed examples.

One source frame helps, but it doesn't settle how the edge behaves when you put the bezel back in the nest. I'd hate to perfect a crop that only works for one lucky seating.

Has the acceptance decision been made before you start that reseating comparison?

Quality accepted the line against the cosmetic examples, and we've retained the bezel with that decision; the reseating comparison is arranged but has not run yet.

Include a known rejected edge in the same comparison. Otherwise you may find a presentation that hides both an acceptable line and a defect you need to see.

We tried both pieces and the rejected edge disappears in one seating too, so the fixture and lighting need more work; thank you, Farah, that second piece stopped us celebrating too early.

There it is. Keep that bad seating in the comparison set; someone will suggest the same nice-looking view again once the awkward sample goes back in its drawer.