Samples retained. Labels came from images only. I've asked quality for the physical decisions.

Keep the original images with pending labels while quality assesses them, rather than discarding the evidence that exposed the uncertainty

Do the filenames let you match each picture to one actual plate? If several views were treated as separate examples, you may need to repair the grouping before you can use quality's decisions.

Most match. Two images have no reliable sample link; I've marked them unsuitable as reference targets.

Keep them as background, with that limitation. For the identified plates, record lighting, exposure and pose in the new comparisons. Those are different possible sources of image change.

Who will maintain the physical reference set after this exercise, so a damaged or replaced plate does not silently retain an old acceptance label?

Quality owns the retained set. Their assessment found accepted plates and genuine scratches; disputed plates stay separate.

Have you repeated presentation yet, or only settled the part labels?

Labels first. Repeated presentation is next. No threshold work started.

When you repeat it, keep the same identified plate in the comparison and include removing and replacing it. Otherwise several still images can hide the part of the workflow that causes the difference.

Keep imaging settings fixed for the first seating comparison. Change light later in a separate comparison. It is much easier to understand which intervention helped that way.

That separation is useful, but the final arrangement still needs a combined check over the expected range of parts and presentation conditions

Who makes the call when a future image remains doubtful?

Quality is defining that route. Repeated seating gives two positions in the temporary holder. Tooling is investigating.

Can the operator tell which position the plate has taken without interpreting the glare? A reliable holder is the first issue, but the eventual loading instruction also needs a usable visual reference.

Don't add an instruction telling them to push until the image looks right. Ask tooling to establish the intended contacts and loading path. The image is what you're investigating, not a reliable seating gauge yet.

Tooling proposed a holder revision with clearer seating. Not built or tested yet. Original image conditions retained for comparison.

Keep the genuine scratched plates in that comparison, because reduced glare on accepted examples alone cannot establish defect visibility

Included. Thanks, Harish; separating the holder and lighting changes gave tooling a clearer question to investigate.