I cannot select inspection references from these trim-plate labels

Quality found one fine physical line, but neither image arrow points to it. The line is accepted under our finish requirement. Both supposed scratch labels came from image-only judgements.

So the arrows are wrong before the threshold gets a turn.

We once preserved every original image but forgot that our teaching captions contained separate verdicts. Check those too when you correct this plate. The next person may read the slide rather than your reference index.

Do you have a rejected plate as well?

Yes. Quality has identified a second plate with an unacceptable scratch. We will use that and the accepted lined plate together. The old arrows and captions are withdrawn.

The accepted physical line is a useful awkward example. Otherwise people can learn that every visible line must be rejected, even after you have sorted out the moving reflections.

Can both plates be seated by the ordinary loader without individual adjustment to improve their appearance? That is worth checking before selecting the images.

And without putting identification labels under the contacts.

The loader has repeated both in the current holder. Seating is consistent. IDs are on their storage sleeves, with each capture linked at loading, not on a contact surface.

Does the real reject stay visible?

Not consistently under the first lighting arrangement. The engineer has a second arrangement that shows it more clearly, but it also makes the accepted line stronger. That is why we kept both.

Good example of the reference set earning its keep. Which distinction is quality asking the inspection to make between those two physical lines?

Ravi's question matters. Better visibility does not necessarily make an acceptance boundary easy to measure, particularly if the current finish judgement includes more than the image captures.

Quality distinguishes the fine brushing-related line from the deeper scratch using the agreed physical viewing method. The camera engineer has not shown that the proposed images capture enough of that distinction.

Then don't label the camera ready for the lesson.

Could the lesson be about the unresolved comparison instead?

That is what we chose. Students will compare the identified samples and captures without being told the camera can already decide acceptance. The automated inspection exercise remains deferred.

I'd rather attend that lesson than tune a number until it agrees with two bad arrows.

Keep the agreed physical judgements available during it. Students can question the image evidence without inadvertently reopening the sample identities you have just established.