Step-by-step assembly playback could show an arriving part crossing an earlier one while the final pose was clear. The fix checks the actual insertion path from offset to seat against parts in install order, bound to hashes of the guide, model and meshes, and missing or stale evidence fails visibly. A follow-up, #517, says the real collection now has 233 instances and can need 27,028 ordered pair checks.
Recurring agent failure issue in cad-agent (Valpatel)
What engineers report
A recurring agent failure problem in cad-agent (Valpatel) that mostly hits the hardware engineer.
Failure pattern: validation / failure
Analyst note
From the Grok pain-radar digests, Nov 10-11 2026 (Nov 10 8:41 AM ET through Nov 11 7:41 AM ET digests plus the Haitch editions). Verified live 2026-11-11: issue/PR read; title and state confirmed.
What it costs a team
Modeled annual cost
$30,000
For a team of 5
Prototype delay
3 days
Modeled schedule slip per occurrence
Time lost
240 min
Per engineer per month, about 1 incidents
How the cost is modeled
240 min/month × 5 engineers × 12 months = 240 hours × $125/hour = $30,000
A modeled estimate with medium confidence. Swap in your own team size and loaded rate to size it for your team.
Engineering tools that address it
No engineering tool profile lists this error yet. Browse tools that connect to cad-agent (Valpatel).
Evidence
More recurring errors in cad-agent (Valpatel)
- An agent-built robot turret rendered its ammunition outside the hopper and its subassemblies unconnected, and per-part FEA and a clash sweep caught neither problem.MCAD · impact 74
- Before this, the fuse board had 6 hand-written footprints and only 7 of 43 footprints had a 3D model. The new parts library builds footprints and STEP bodies from datasheet numbers, each tied to the page it came from, with a registration check confirming the body sits on the board, pins go through their pads' holes and pin 1 lands on the right pad. After the change, 37 of 43 parts have a body, and the other 6 are waived with reasons.ECAD · impact 74
- For a McMaster heat-set insert with a 4.0 mm maker's hole, the agent's hole_design module drew 4.20 mm, while fit_coupons centred its test sweep on 4.25 mm. The author traced the right number to the CNC Kitchen test data already cited in the repo and fixed the coupon. Two modules in one agent, same part, two numbers. This is the shared-source-of-truth problem at the scale of a single repo.MCAD · impact 74
- OCCT Common and Cut booleans missed a 4 mm sphere pushed 0.03 to 0.1 mm into a cone wall at 5 of 50 positions, reading 0 mm3 outside, so the volume cross-check could not see it.MCAD · impact 74
- OCCT's boolean can miss a shallow crossing between two nearly tangent curved faces, and in a ball-through-cone sweep 24 of 500 cases were misread as contains or touching, hiding a clash.MCAD · impact 74
- On cad-agent excavator project, fusing nine nearly coincident swept solids in one call gave an invalid shape. The invalid tool then cut nothing from the boom, and only a range-of-motion collision check caught it, 14.7 cm3 of overlap. A cone meeting a pin hole raised Null TopoDS_Shape until the tool was nudged 0.01 mm, and a sub-micron split vertex made a valid solid fail the watertight export.MCAD · impact 74
Sourced from public community reports. Costs and delays are modeled estimates, not measurements. Know a better source or a fix? Tell us at hello@productflo.io.