Three bezels are retained with image links. None was formally rejected; the captions came from our team comparing pictures on a laptop.

Who wrote the captions, Omar? I would want the reason recorded, even if it turns out to be a guess, so the lesson doesn't quietly acquire three supposed inspector decisions that nobody actually made.

I wrote two, colleague wrote one. We pointed at the brightest lines. I've marked that origin plainly and asked quality to judge the physical pieces.

Do the image links identify the same region across presentations, or only the same bezel? Those give you different levels of confidence when someone circles a bright patch that has moved away from the original location.

Use stable physical features to map the regions, not the highlight itself. Keep the full view with any crop. That lets an imaging reviewer compare a specific surface area under different conditions without pretending the reflection is a fixed feature.

Imran, yes. Our other cover comparison still has accepted surfaces making strong glare. A good physical reference can produce an awkward image; that awkwardness is part of the test, not a reason to discard the reference.

Harish, the bezel links are definite, but the old circles follow the bright lines. I can map the actual regions against the locating holes instead. That annotation work is in progress.

Reece, that's exactly the distinction the training notes need. I'd hate the next person to learn accepted means an image with no bright patches, then be handed real reflective parts.

Quality accepted all three under the current appearance requirement. No reject scratch on any of them. So our folder has false defect captions, not three confirmed defective references.

Do you have any physically confirmed reject examples outside those three? Correcting the captions gives you accepted material, but it doesn't establish what the camera must reliably find on a damaged bezel.

Keep the three as accepted references and separate that missing reject coverage. A test that accepts all three may still miss every required defect. You need representative examples and a defined requirement before interpreting a clean result as detection capability.

No retained reject bezel available here. Quality is looking for traceable examples from the current process. I don't want us inventing a scratch label just to fill the reject column again.

Then the reference set is useful but incomplete. You can study false rejections on the accepted surfaces while making no claim about defect detection.

Will the team actually keep that qualification on the results page, not only in your email? Someone seeing three green results could call the trial finished without reading this discussion.

Yes. Results page separates accepted-surface responses from untested defect coverage. No overall pass box. Thanks Yasmin, that is clearer than burying the limitation under the sample photographs.

And have you recorded the presentation and lighting for the accepted-surface comparison? Otherwise even that narrower result will be difficult for the next person to reproduce.

Holder position, camera settings and lamp arrangement are recorded for the planned comparison. The old images stay as diagnostic history because those conditions weren't recorded then. No new comparison results yet.

That preserves the useful distinction: old evidence exposed a labelling error, and the new controlled comparison will test a defined question. Neither needs to impersonate the missing reject study.

Could you post what traceable reject examples quality eventually finds? Particularly whether they represent the requirement people were trying to describe with those bright-line captions.