It specifies the viewing method. We didn't use that method to label the image set; the meeting just circled bright lines.

That gives you a concrete correction to make. Retain those annotations as provisional, associate them with identified bezels and obtain the physical judgements. It is not yet evidence that every circled line is harmless.

Reece, can Carla start recording seating conditions while the physical review is pending? Otherwise those images may acquire better labels but still lack capture context.

Yes. I meant the judged references are needed to assess visibility, not that all observation must wait. Keep those two strands moving together.

I'll collect the holder and presentation records while quality reviews the retained bezels. Thanks, that avoids booking two separate rounds just to recover the context.

Keep a few independently selected examples outside tuning when you get that far. Otherwise the same small set can become both your target and the evidence that you hit it.

Physical classifications are back. The image labels are corrected, but one rejected feature isn't consistently visible across repeat placements. Threshold work remains on hold.

Which capture condition changes its visibility? Preserve the rejected reference while reviewing that, rather than replace it with an easier defect to finish the exercise.