Nobody yet. Two retained plates, both identifiable, but the scratch judgement came from viewing the photos. I've stopped treating those labels as reference answers.

What viewing and cleaning conditions does the appearance requirement use for the physical decision? I would settle that with quality, then preserve the identified samples for the camera comparison rather than let each person judge under their favourite light.

And keep accepted finish examples. A scratch-only set won't show false rejects.

Quality is taking both retained plates and defining the physical viewing method. No verdict yet. Accepted finish variation and confirmed defects are in the requested reference scope.

Our camera trial improved when we changed one thing at a time instead of moving the part and the lamp together. I'd keep the observed presentation differences as test cases after the physical decisions are settled, so you can see what any proposed change actually fixes.