Quality has the appearance requirement. The labels came from our image review meeting, without the parts on the table.

Were those meant as final labels or working notes? I'd settle that before telling the whole team they have been training on errors.

Working notes can still become threshold targets by accident. Mark them unverified now, whatever the meeting intended

They were provisional, but the tuning spreadsheet omitted that word. I've restored it. We haven't tuned from it yet.

That saves a confusing handover. I'd pair each provisional image with an identified cover and the presentation conditions, so quality can explain what it actually judged.

And preserve a separate unknown category where the surface itself remains disputed. Forcing every example into pass or fail can make a tidy spreadsheet without giving you usable truth.

Quality can review the physical covers next week. I'm collecting the image-to-cover links first; a few images have only a tray number.

Thanks for catching that missing word before tuning. Leave the tray-only images unassigned unless someone can establish the actual cover; resemblance isn't enough.

Once the references are judged, test repeat presentation with them. Reliable labels and repeatable visibility are separate jobs. Is anybody owning the fixture and lighting records?