They haven't judged these plates yet. I've kept the existing captions marked unverified, separate from their IDs.

Keep the brushing direction identifiable in each view. I would use an existing edge or hole as the orientation reference, so nobody has to infer it from the bright band itself.

Quality accepted the bright brushed plate and rejected a retained cross-grain score. Both physical decisions now have locations.

Try to make that accepted plate look bad in the comparison too. Not by editing the picture, I mean include its permitted presentation variation. A threshold that only survives the neatest example is no use at changeover.

Nora, agreed, and the scored plate needs that same variation so removing the glare does not also remove the thing Leo needs to detect.

The first comparison mixed brushing directions. Our holder permits a half-turn which the loading instruction never distinguished. Imaging has compared both: the score disappears in one orientation under the current light. So a threshold isn't the first decision after all.

Could the part still arrive in either orientation in the real workflow? If yes, writing an arrow on the reference sheet will not make the unwanted presentation impossible.

Yes, either can arrive. We're assessing imaging for both rather than relying on an arrow; no candidate has passed both yet.

Keep both orientations in the acceptance set, then. Your physical reference decisions are useful now; the imaging result still has an explicit failed condition instead of a conveniently omitted picture.