Keep a known rejected feature in the lighting comparison once quality identifies one; removing every visible line would be a poor inspection improvement

Quality examined the retained set. One marked bezel has a real scratch. Two bright-streak examples were accepted surfaces. Those decisions are now linked to the original frames.

How are you keeping the reference bezels between sessions? A changed surface could leave the next image set attached to an old decision.

Individual identified storage, with quality checking condition before reuse. We're varying presentation separately from lighting in the next capture session.

What happens to the old scratch labels? Don't leave two contradictory sets for the next person

Jane, keep the original annotations for traceability but remove them from the current tuning set. Deleting history and continuing to use bad labels are not the only choices.

Old annotations are archived, not current training references. New set points to quality's decisions and the capture settings.

Yes, archived was what I meant. Not erased. Thanks Bruno

Capture comparison done. One light position reduced the accepted-surface streaks but also hid the confirmed scratch at one allowed presentation. We rejected that position. The other candidate kept the scratch visible in this small set.

Does the retained set include the range of finish you expect from production? I don't know that range, but one scratched bezel and two accepted ones sound like a starting comparison.

Starting comparison only. Quality is choosing more finish examples. We have corrected the labels and rejected one lighting option; no inspection threshold accepted yet.

Record why that light position was rejected, including the scratch view it lost, so nobody brings it back based only on the cleaner accepted images