Separate the guessed annotations from quality's decisions right away. Otherwise somebody will reuse the pictures while you're arranging the review. The distinction needs to travel with the image, not sit in a message nobody opens.

The faceplates are retained, but two example photographs have no part identity. I've moved those out of the tuning set. Quality has the identified parts and the disputed regions; no physical decisions back yet.

Keep those two photographs for diagnosis rather than deleting them; unlinked isn't the same as useless, just not a trustworthy labelled reference.

Who loads the holder for the repeated pictures?

The vision engineer made the examples. I'm asking our normal loader to join the comparison, with the placement method recorded, so we can see what variation the actual handover leaves.

Give both people the same method without having the designer quietly adjust every placement. Record any assistance as an observation. It can show that the holder or instruction needs work, even if the resulting picture looks excellent after the adjustment.

Has quality found an actual defect example yet? An accepted shiny faceplate can teach you plenty about glare while teaching you nothing about whether you still see a faint scratch.

Daniel, it can tell you what not to call a defect, though I agree it cannot establish detection by itself.

Quality returned an accepted faceplate with the broad travelling highlight and a different plate with an identified faint score. They also left one disputed area undecided. That last one is not a tuning target.

David, yes, I overstated that. The accepted one matters for false rejects. I wanted the comparison to ask both questions, and Omar now has parts for that.

What did the normal loader's placements show?

The loader could seat the faceplate in two slightly different resting positions that both appeared to follow our instruction. We stopped treating those pictures as a lighting-only comparison and sent the contact observations to the holder designer.

That is worth separating. The holder review should define the intended contacts and normal loading access; then repeat with the identified accepted and defective parts. A firmer-looking seat isn't sufficient if the new orientation conceals the faint score.

Are you changing the lamp as well during that holder comparison? It would make the source of any improvement harder to explain.

No lamp change in the first holder comparison. The designer has clarified the intended contact arrangement, and we are recording whether people can achieve it without help before testing lighting alternatives separately.

Keep the undecided physical area out of the pass-fail score when you get there. It can stay on the quality review list without forcing a convenient label to finish the spreadsheet.

The revised loading instruction removed the two-position ambiguity in our supervised repeats. Both identified surface features remain visible in those captures, but we have not done the wider evaluation. The undecided region is still with quality.

Reserve independent parts for that wider evaluation once you settle the imaging candidate. Repeating the same diagnostic plates is useful development work, not an independent test of the finished inspection.

I've added that separation to the commissioning plan. Thanks, Yasmin, bringing the normal loader in found something the engineer's own pictures never showed. The cell remains uncommissioned while the reference boundary and wider evaluation are unfinished.