Yes. Reverse has a sample mark. All three pictures are the same bezel.

Who put the scratch labels on those pictures, Rob?

Me, after a planning meeting. Nobody inspected the bezel. That's why I'm trying to stop this becoming training material.

Get quality to examine it against the appearance requirement and mark any actual defect location. Keep your existing images as an example of the uncertainty, but exclude the unverified labels from threshold tuning.

Has that folder already been copied for training? Worth checking now. Correcting the one in front of you doesn't reach the copy someone put on a laptop for the commissioning visit.

One shared copy. Asked its owner to mark the examples unverified too.

Would you keep an accepted bezel beside any reject quality identifies? I mean a physically agreed accepted piece, so the image comparison includes both things you need the camera to distinguish.

Quality accepted this one. Bright stripe not a scratch. They're selecting a real reject for comparison.

Are the original capture settings available, or did the team adjust them without keeping notes?

Exposure logged. Light angle changed by hand, no notes. Three flattering portraits, no repeatable setup.

Then repeat a small comparison with positions and settings fixed and recorded. Once that is stable, remove and replace each part. You want to see what seating contributes before collecting hundreds more images.

Also show the whole bezel in the inspection record. A cropped bright line can become very persuasive when nobody can locate it on the actual part.

Did quality find a suitable reject, Rob? Interested in whether the defect is near the same reflective edge or somewhere easier to image.

Small nick near the edge. Identified and recorded. Harder to see than my imaginary scratch, naturally.

Use that nick as the target feature and check the accepted edge alongside it. A setting that hides all bright edges may make the accepted sample look clean while hiding the reject too.

What does the repeated seating show so far?

Accepted bezel can settle at two angles in the holder. Fixture person has measured the contact difference. Revised seating proposed, not tested yet.

Has the shared training copy been updated? That earlier question is still hanging while the physical investigation moves on.

Yes. Owner confirmed its old labels are withdrawn. New reference images will wait for the holder check. Thanks Nora for chasing the duplicate copy.