Who should decide which marks on the actual trim plates count as defects for my FR5 lesson? The station is still at the planning stage, yet the example folder has already promoted a travelling bright streak to a scratch; I am not keen to have students tuning thresholds around that decision.
The person responsible for the trim plate's appearance acceptance should judge the sample under agreed viewing conditions. Keep the uncertain images outside the approved training set until then, with the original sample identity still attached
Our finishing inspector has the appearance sheet and will examine the retained plates, but the camera folder only has filenames, so I am matching those to the physical labels first
Can you establish those matches reliably, or were the filenames assigned after the plates had already been mixed? I would not attach a confident sample identity based only on similar-looking reflections.
The original capture sheet links most filenames to plate labels, but two images have no entry; those two are excluded from the reference selection rather than guessed into place
That leaves a usable starting set. Have the inspector include a known unacceptable mark as well as accepted surfaces, so improving consistency does not simply mean making every plate look equally featureless
We now have labelled accepted plates, a rejected edge rub and one uncertain sample retained separately, with the appearance decisions attached; the imaging trial still needs to show that rub reliably
Did the imaging comparison make the edge rub visible under repeatable presentation? I would ask that before spending more time on a threshold for the larger moving streak.