Repeated with fixed settings. The accepted streak is obvious; the actual scratch is still barely visible.

Is the scratch in the part of the image you intend to inspect? A full photo can contain it while the inspection crop misses it.

And retain the settings with those files, so the next person can repeat this baseline instead of finding a fresh automatic setting

Don't delete that disappointing pair, Rob. It's the reason you're considering different lighting, not a failed photo session to hide.

Victor, the scratch is inside the intended inspection area. Naomi, the baseline settings are saved beside both originals.

Can the second loader reproduce the intended seat with the actual holder, rather than having someone arrange the plate for the photograph?

Second loader repeats the seat. Imaging has proposed a different light position for the same sample pair.

Did it improve the scratch view, or just move the acceptable streak somewhere less annoying?

Scratch is clearer. Accepted streak remains visible but distinct under quality's comparison. Broader sample set next.

That's promising, but I would bring in a weak scratch, not just more obvious ones. We nearly convinced ourselves with an example you could identify from across the bench. The troublesome accepted finish needs company too; one clean control doesn't represent every acceptable surface.

Who replaces that light position if it gets disturbed during cleaning?

Naomi's restoration question should reach the holder and camera as well, although each may need a different check.

Agreed. A pencil outline round the lamp is not the entire inspection setup, tempting though that budget is.

Maintenance owns restoration checks. Expanded comparison found one faint reject still unreliable; we have not selected a threshold.

Is that faint reject visible enough in the raw image for a person to judge consistently? I'd ask that before someone spends a week adjusting the threshold around missing information.