Three retained. Two image links are certain. The third group is only labelled cover samples.

Keep the third group unassigned while checking any original capture notes. You can retain its evidence of an imaging problem without inventing which physical cover it represents. Have quality judge the two linked pieces against the appearance requirement independently of those captions.

And map the actual regions between views, not whichever bright line the cursor follows.

Quality accepts both linked covers. The travelling lines are not reject scratches. No confirmed defect reference available yet.

Keep those accepted reflective surfaces in the comparison rather than choosing duller examples that make the camera's life easier. On our own faceplate work, retaining accepted pieces alongside a rejected one made the different questions much clearer; you still need the appropriate defect examples for yours.

Who owns obtaining those defect references, Sarah? The current two cannot test detection by themselves.

Quality is sourcing identified examples. Imaging is comparing the accepted covers under recorded seating and light conditions meanwhile.

What does that accepted-cover comparison actually show? I'd keep false rejection separate from the missing detection study, so an improvement in one cannot quietly stand in for the other.

One proposed light position reduces false scratch responses in repeat views of both. Detection coverage remains untested.

That is useful within its scope. Keep the previous strong-band views too, because the next reviewer needs to see what changed and why the new arrangement was considered, not only the prettier results.

Don't select the final threshold until the required defects have had their say too.