Quality has the retained pieces now. Two image filenames cannot be linked to a specific bezel, so I've separated those from the candidate references

Keep those unlinked frames as evidence of the imaging problem. They can show a difficult reflection without becoming examples whose physical defect status you claim to know.

Who is keeping the parts after quality examines them? I've watched useful sample sets drift back into ordinary stock when nobody knew why the pieces were being saved.

Quality owns their storage. They confirmed a scratch on one identifiable bezel and accepted two others with troublesome reflections. The unlinked files remain unassigned

Will the capture comparison include repeated loading by the normal operator, not only a carefully placed reference that never moves? That seems important for a robot-presented view later.

And keep lighting changes separate from presentation changes initially, so an improved image has an explainable cause.

First comparison complete. The confirmed scratch stays visible at the intended pose, but glare hides it in one allowed seating variation. That variation is now in the fixture and imaging review rather than excluded from the results

Does that mean you are narrowing the permitted seating, or changing the imaging? I would want the loading instruction to reflect whichever decision is actually made.

Neither selected yet. Fixture owner and imaging lead are comparing both options. Thanks Thomas Adams; the instruction is on their review list too. We have corrected reference labels, not a validated inspection setup

Keep the troublesome view in that comparison. Removing it would make the results tidier without answering how the allowed condition will be handled.