The example folder labels a bright line as a scratch, but the line changes position when the bezel is reseated. We still have the labelled bezel. I would like inspection to judge it before the imaging trial continues. What should go into the comparison so we do not simply choose the prettiest lighting?
Keep the labelled original, then compare that same identified bezel with inspection's physical judgement; a moving bright line is not a reliable defect location.
Inspection accepted that bezel. There is a separate rejected example with a shallow scuff near the mounting hole. Both are now identified and kept out of production. The old bright-line captions have been withdrawn.
Use both in each lighting comparison. On a trial here we made the accepted part look beautifully quiet, then realised the same arrangement hid the shallow reject. Keep full views as well as crops so you can see seating and orientation. I'd start with repeated bench loading before spending the robot session discovering that two people seat the bezel differently.
Include the person who will load it, because a good image from the trial engineer's favourite hand position may add an awkward preparation job for someone else
The first bench comparison found exactly that tradeoff. One light position suppresses the streak but makes the scuff faint. A second view shows it clearly. The bezel stayed against the same contacts, with loading repeated by our operator.
Inspection added accepted bezels with the ordinary finish variation and another rejected scuff in a different place. Both proposed views were captured for each. We are keeping the two-view candidate; the single quiet-looking view missed a rejected example in the comparison.