I have stopped using the arrows as defect references. On our retained polished covers, the bright band crosses different features after reseating. Quality has identified a real scratch elsewhere. Which examples belong in the camera comparison?
Use the physical accepted cover and the scratched one, both identified. Keep the bad arrows somewhere outside the reference set so you can explain why they changed, but don't leave them where somebody can reuse them as labels
Don't delete all the awkward images. They show the presentations that fooled the current labelling. I'd want those in a challenge set, with corrected labels.
Ben, agreed, but the correct label needs quality's judgement on the actual cover. We had a wrong arrow on a genuinely rejected bezel. Removing the arrow did not make the bezel acceptable.
Amy, full images are retained. Ben and Nora, quality calls one cover rejected for the small edge scratch, not the travelling band. The accepted cover has the band too.
Emma's distinction matters. A deliberately tilted photo can be useful for finding a reflection, but I'd keep it separate from the conditions you expect the loader to produce
Beth, the scratch is faint but visible in one view. Omar, the locating hole identifies its edge. Emma, two photos were deliberately tilted; the others came from ordinary reseating.
Then I'd split deliberate tilts from ordinary reseats in the comparison. Otherwise somebody will tune against an impossible loading position and call the normal ones easy.
We are comparing two different aims here. The ordinary reseats tell Grace whether the image stays useful during loading. The tilted images help explain why the band moved. I would retain both, but not let either stand in for seeing that edge scratch reliably.
Camera engineer compared the two sets with quality. They removed the band labels and rejected the current lighting candidate because the edge scratch disappears on an ordinary reseat. No threshold trial with that candidate.
Another view is being considered. We have not selected it. Thanks, Emma, separating the deliberately tilted images stopped us arguing about a loading condition we do not use.