A text-to-cad model shows a shop what to make but not how exact it has to be: hole size limits, position and flatness sit only on a drawing. The PR writes them as AP242 semantic PMI into the STEP, reads each STEP back, refuses one that lost a requirement, and lists faces with no requirement instead of filling them in.
Recurring agent failure issue in text-to-cad (earthtojake)
What engineers report
A recurring agent failure problem in text-to-cad (earthtojake) that mostly hits the mechanical designer.
Failure pattern: validation / failure
Analyst note
From the Grok pain-radar digests, Nov 9-10 2026 (Nov 9 7:41 AM ET through Nov 10 7:41 AM ET digests plus the Haitch editions). Verified live 2026-11-10: 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 text-to-cad (earthtojake).
Evidence
More recurring errors in text-to-cad (earthtojake)
- A sheet-metal DFM check passed a flat pattern that was 0.4 mm short each way, with nothing said. An 80 x 40 x 2 mm plate with 0.2 mm rounds on both skins unfolded to 79.6 x 39.6 and the sheet check said OK. With 0.5 mm rounds it failed with the flat pattern holds 97 percent of the part's material, and the unfold function raised an error. The unfold used the skin outline, not the cut outline, and hexagons and chamfered outlines do it too.MCAD · impact 74
- After moving to OCP 8, strict STEP volume comparison failed on an F-14 empennage sample on both versions and a direct build123d STEP import produced two invalid solids, while the project's own reader called every prototype valid.MCAD · impact 74
- The tool meshes every part at 0.25% of the mean bounding-box side, and the viewer and STL/3MF exports share that mesh, "configurable nowhere". For an 85 x 94.5 x 75 mm hopper modelled with a 0.2 mm offset around a grinder body, the export's deflection is about 0.21 mm, roughly the entire clearance. The reporter measured it against their own BRepMesh run at 0.01 mm. A setting chosen for the viewer quietly decided the manufacturing file.MCAD · impact 74
- A new four-way failure classifier duplicated what telemetry already decides when it counts a crash. They dropped it and reused telemetry's rule so the Report Issue button and crash counts can't disagree. Small, but the same pattern as two parts of one system computing the same fact.MCAD · impact 66
- cad_http was listed for the agent so it could flip telemetry; now it is app-only, and a call without the page's capability header is refused before routing, while the agent keeps cad_open/show/view/screenshot and export tools. The tunnel shared between page and agent was the wrong trust boundary.MCAD · impact 66
- OCCT Booleans on two lofts sharing a section return an empty fuse, a zero common, or a cut bigger than the original body.MCAD · impact 66
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.