Two retained plates have conflicting image labels. Neither has an agreed physical judgement yet.

Keep both out of the tuning references until that is settled. You can still use their images to investigate presentation sensitivity. Uncertain examples are useful, just not as unquestioned answers

What would an operator be told when the same plate looks rejected in one view and accepted in another? I'd include that uncertainty in the training notes now, rather than let the brightest image become the unofficial standard.

There isn't guidance yet, Yasmin. The example folder is being mistaken for an acceptance standard.

Then label the folder's purpose. Ours collected reject in its name and suddenly everyone treated every photograph inside as a verdict. Very efficient way to manufacture certainty.

Chen's distinction helps, but keep the per-plate uncertainty visible too. Folders get copied and renamed. The part identity and physical decision should travel with the reference, not depend on remembering its original location

I've separated investigation images from references, with the two plates explicitly awaiting quality judgement.

Keep both views available to the trainer, Sara, once that review explains them. Seeing why a highlight wasn't the defect would help me more than being handed only the final tidy example and told to remember it.