FR5 planning images have moving streaks labelled scratches. No commissioned station yet. I want the physical references sorted before the camera gets taught our guesses; what's missing from the handover?
Retained covers, identified regions and a physical decision under the appearance requirement. Keep the different presentations linked. The movement of a streak is worth investigating, but it does not tell you whether the surrounding surface is acceptable.
Recover the full views before treating those crops as reference evidence. A crop can be perfectly sharp and still be a picture of an unidentified bit of cover. I'd rather repeat an uncertain capture than let the next tutor guess its physical location.
Full views recovered for one session. The other crops have no reliable parent image, so that session is excluded from confirmed references, not silently assigned to whichever cover looks nearest.
That narrows the provenance problem. Quality still needs to decide the retained surfaces, and the imaging comparison needs known regions rather than the most dramatic crop. No reason to tune thresholds from this set yet.