I've removed guessed labels from a faceplate tuning set before, while preserving the disputed images. This planned FR5 polished-cover station has the same moving-streak problem. How should I organise physical decisions and repeatable imaging together so we don't simply build another uncertain set?
Keep one identity for each physical cover across inspection and capture records. Have the appearance owner judge the surface under the agreed method, while recording the presentation and imaging conditions independently of that judgement.
We retained the selected covers. I can establish those links; the wider image folder has weaker identities and isn't going into tuning just because it is larger.
Make the capture comparison repeatable before expanding it. Same identified covers, declared holder and lighting, then deliberate presentation cases. Changing lamp, seating and exposure together leaves you with another argument about what helped.
Keep that as an observation about the current method, not a proven cause. Physical surface decisions can proceed while the holder review establishes which presentation differences to compare.
They will. Quality has classified the selected covers and left one faint feature unresolved. It stays identified outside the tuning targets, not discarded.
Use both confirmed acceptable and rejected references in that revised comparison. Improved seating is useful only if the required surface features remain visible across the intended presentation range.
Keep validation selection separate from the images used to tune. Otherwise the team may only prove it can reproduce decisions on its familiar small set.
Quality is selecting that separately. Thanks Jo, the regular loader is reviewing the proposed seating arrangement, so its explanation won't depend on me being beside the bench.