I was collecting training pictures of our brushed trim plates and a horrible bright line disappeared when I moved the spare-parts rack behind the camera. We're planning UR5e presentation to that camera. I now have a folder of rejects that might mostly be reflections, and I'll be the one explaining it to evening shift.
Have you still got the actual plates that produced those pictures? Start with a few identifiable pieces and have quality examine them. Keep their judgement separate from the bright-line labels. Otherwise you could build a beautifully consistent test for the wrong feature.
I kept four of them in numbered sleeves. Quality found a real scuff on one, but it runs across the brush direction and isn't where our bright line was. The other three are acceptable to them.
Use those as an initial reference set, with quality's locations marked on a separate image. Photograph each repeatedly in your intended presentation. Include the surrounding rack position in your setup notes; it has already shown that the environment changes the result.
We've already done a static repeat set this morning with a screen around the station. The big false line is gone. The real scuff still isn't consistently visible, so I'm keeping this out of the operator pass/fail instructions for now.
That is a useful stopping point. Suppressing a reflection has not demonstrated scuff detection. Give the vision person the accepted pieces and the scuffed piece together, and ask for an imaging trial before anybody chooses the threshold.