my bracket only looks square while it's trapped

I had two support sets with the same bench nickname. The drawing revision helped us find the right one. Is your check arrangement actually identified?

Anna, drawing P, measured released on the inspection support named on that sheet; Mina, yes there's an asset label and it matches, I've checked that rather than trusting the nickname this time

Then you've got a usable before-and-after comparison if you keep the part identities together. For the fixture model I'd show where each piece sits before the clamps close, otherwise you can accidentally draw the corrected shape as the starting shape

Why are we blaming the weld already? Could the gusset be wrong before anyone strikes an arc? I would want that answered before paying for a different welding sequence.

We mixed two almost identical gussets once. Awful afternoon. Worth looking at the actual edges.

The fabricator has a fit-up photo from the failed bracket and the gusset isn't against the rear stop before clamping, I'd missed that because the later photo is all closed up (thanks Sam, wrong photo was doing all the talking)

Luca, show the intended seating too. A gap in one photo isn't automatically the design.

Yes, both. I'd label the photo as the observed fit-up and put the designer's intended contacts beside it, so nobody models that gap as something we're meant to preserve

We found a locator from an earlier assembly on a similar fixture. A comparison of the fitted stop with its drawing was cheaper than our first proposed experiment.

Tooling found an old stop still fitted behind the gusset, its number belongs to the earlier bracket and the current assembly calls for a different face position, so the fabricator wasn't imagining the awkward fit

Have they checked first release after correcting that assembly? Keep any later correction separate. That will tell your estimator whether the old allowance still describes this job.

Three identified brackets through the corrected fixture now, all passed the P check at first release with no correction, and the estimator has replaced the old assumed correction time with those actual observations plus a note that it's only three parts; I can finally explain the price change without pretending the first failures never happened

Who checks that stop after a changeover? Put a name against it while everyone remembers.