our scratch folder contains the same plate twice

Separate folder, then. I was about to delete the doubtful ones entirely.

The capture sheet has front and back photos under each plate number. These two still have file names that join to that sheet. The others were renamed for a presentation.

Could the presentation file have the original pictures embedded? I mean the full pictures before cropping, not necessarily the original filenames.

Worth looking, but Pavel still needs the link to the physical plate; a larger mystery picture stays a mystery.

Theo, yes, the two identified captures have full frames on the capture machine. Same plate number and back photo in both entries. Its condition still needs checking.

Cheers, that joins it up. I wouldn't bin the unlinked ones either, but please keep them out of whatever somebody later exports as the labelled set

Sam, I agree about identity. The full frame helped me find a crop that had been taken from the wrong end of a part. It didn't tell me whether the surface was acceptable.

Noah's suggestion recovered wider pictures from the slides, but no dependable plate numbers. Those are now in a separate lighting discussion folder. Nothing from that folder goes into the labelled export.

What did the actual plate look like?

Glad the slides yielded something. Were both supposedly opposite labels attached to the same area of the plate, or were they describing different areas?

Same area, Noah. Quality checked the identified plate and accepted that area under the agreed physical inspection method. The line in one capture is conspicuous, which is why somebody promoted it to scratch without looking.

That accepted example is useful. Have you got a physically rejected mark in the same region to compare? My nice clean examples stopped being informative once we added a different finish.

Who gets to promote an image label now? Because changing the folder structure won't stop the next person doing exactly that.

Quality owns the surface decision and I've taken ownership of joining that decision to the image files. Nobody is meant to judge a part from the slide thumbnails. That instruction was missing before.

Theo, we have a rejected plate with a physical mark in that region. Its image pair is being made under the same recorded setup; I won't quietly borrow one from the old mystery pile.

Use that pair before shopping for another camera, especially if the current views cannot keep the accepted streak and rejected feature apart.

The new pair is back. Our present view makes the accepted streak stronger than the real rejected feature, so the team dropped the proposed brightness cutoff. We're trying a revised lighting arrangement on identified plates; the image folders are repaired, the inspection method isn't finished.