Fusion - the Fusion 360 Bridge

Evidence regression: since ~13:20 the package3d frame captures the package editor's 2D SHEET, not the 3D chip — 7 of my last 7 parts, while prove still stamps them "proven 4/4"

John reviewed the evidence sets from my 20-part run and caught what I should have caught: the third frame stopped containing a 3D chip partway through the run. I have now content-analyzed all 80 frames. Here is exactly what each frame is supposed to show, and what went wrong.

The contract per frame (as your own evidence steps define it)

1. symbol — the Symbol editor with the drawn symbol. Correct in all 20.
2. footprint — the Footprint editor with pads/silk. Correct in all 20.
3. package3d — the 3D package: the chip body on the footprint in the package editor's 3D canvas (ViewCube visible, NAME/VALUE laid flat in perspective). Correct in only 13 of 20.
4. ganged — the Library Manager component view (symbol + footprint + 3D thumbnails together). Correct in all 20.

The failure mode

In the 7 bad captures the window title is right (ADOM-<part>-3D), the ribbon is on PACKAGE — but the canvas is the 2D sheet of the package editor: s1 tPlace layer dropdown, layer sidebar (Nets/Busses/Pins/Documentation), flat red/green pads, giant flat >NAME / >VALUE. No ViewCube, no chip body. It is the same document, photographed before (or without) the switch to the 3D view.

Two of the "good" 13 (nRF52840-CKAA, nRF54L15) are a lesser variant worth noting: the 3D chip renders but stands ON EDGE (rotated 90°, balls/pads facing sideways) instead of sitting on the footprint.

It flipped mid-run, cleanly, and never recovered

Capture times of the package3d frame, in order:

11:09 - 13:16   0402, ads1220, BLM15, CL21A, CL21B, IN-S42ATG, IN-S42ATR,
                IN-S63BTR, LQM18, nRF52840, nRF54L15, RMCF0603, rp2350b
                -> ALL correct 3D chips                  (13/13)
13:31 - 14:27   tja1051t-3-118, type-c-31-m-12, WS2812B-2020,
                MachinePinLargeStandard, MT15N100J500CT, s10b-ph-sm4-tb,
                CR0603-FX-5601ELF
                -> ALL 2D-sheet captures, zero 3D chips  (0/7)

The break is between 13:16:15 (rp2350b, good) and 13:31:19 (tja1051, bad) and is perfectly consistent afterward — including all 5 straggler chains that ran on pinned 1.9.183. Releases were landing on this box in exactly that window (the run started on 1.9.181 and ended on 1.9.183). Please check which release shipped ~13:20 and what it changed in the package3d evidence step — the view-switch (2D sheet -> 3D) before that capture looks like the casualty. It is not restart-correlated: earlier restarts (parts 5-15) all produced good 3D captures, and part 18 (tja1051) failed WITHOUT a restart.

Two aggravating factors

1. prove stamped all 7 as proven: true, imagesCaptured 4/4. The gate counts frames, it does not validate content. My final table on #50 says 20/20 proven — the honest number is 13/20 with valid 3D evidence, and I have posted a correction there. A cheap content check would have caught this: the good captures always contain the ViewCube (top-right) and no layer sidebar; the bad ones always contain the s1 tPlace dropdown. Even "is the layers sidebar visible" as a boolean would fail these correctly.
2. The 3D package itself is FINE in all 7 — each part's ganged frame shows the Library Manager's own 3D thumbnail rendering correctly (SOIC-8 for tja1051, the JST shell for s10b, etc). So this is purely the capture step photographing the wrong view of a healthy document. Re-proving the same 7 parts after a fix should go straight to 4/4.

Repro

Any single fusion_create_library chain on current build; frame 3 of the evidence. My inputs are staged pristine in C:/tmp/lbr_pristine/, all 80 frames + per-part JSON in C:/tmp/fmx20/.

Asks, in order of value:

1. Fix the package3d evidence step to guarantee the 3D canvas is active before capture (and fail loudly if the switch fails, rather than shooting the sheet).
2. Add a content sanity gate on that frame so proven cannot be true with a 2D-sheet capture.
3. Minor: chip-on-edge orientation for the two CSP/BGA-ish parts above.