Yes, four retained with their image groups. Brush direction is not recorded in the captures; the bright line shifts when presentation changes.

Record brush direction in the new comparison. Do not infer it from the strongest stripe

And let quality judge the actual plates. Four linked image groups are useful, but they don't magically turn four meeting-room opinions into defect references.

Keep the claimed regions identifiable from physical edges or features, then ask quality for the decision under the appearance requirement. Once that is established, the imaging comparison can ask whether accepted finishes and required defects remain distinguishable under your recorded conditions.

Who keeps the plates available while that comparison is being organised?

Thanks for saying where they are. I've seen parts in a folder turn out to mean photographs in a folder, with the actual pieces long gone.

Keep each crop linked to its full view too. Plate identity alone will not locate the region

Full views linked. Quality has not returned the surface decisions yet, so I cannot say which example should be accepted or rejected by the image test.

That is the remaining reference question. Preserve the uncertain material for diagnosis, but leave threshold selection waiting on defined physical examples rather than letting the present labels become correct by repetition.