MECH-027 · Parametric Mechanical Parts · difficulty 5/5

Planetary-gear carrier plate

sha256:2b97cc4d1ef0aa55…

§1Prompt verbatim

Disc Ø 80 mm × 8 mm with: (a) central Ø 12 H7 bore, (b) 3 satellite bores Ø 6 H7 on a 30 mm PCD at 0/120/240°, (c) 6 M3 tapped holes on a 60 mm PCD at 30° offset, depth 6 mm, ISO 261 thread. True-position 0.05 mm to datum A (central bore axis).

§2Ground-truth spec

shells1

watertighttrue

manifoldtrue

acceptance ε±0.05 mm

featuresbore_H7_12, bore_H7_6_x3, thread_M3x6_x6, PCD_30, PCD_60

§3Reference render

canonical reference · drag to orbit, scroll to zoom

Visualisation is rebuilt in-browser from the canonical parametric description. Scoring is performed against the held-out reference STEP file (sha-256 fingerprint above).

§4Per-agent renders

reference + 10 agent outputs · scored against the held-out STEP

vol IoU · BREP · manifold

canonical reference

REFERENCE

canonical · ground truth

1.000100✓

GPT-5 → CadQuery

OpenAI + CadQuery 2.4

0.6939✗

Human Baseline (Mech-E)

n=4 senior engineers

0.6879✓

Claude Opus 4.7 → CadQuery

Anthropic + CadQuery 2.4

Claude Opus 4.7 → OpenSCAD

Anthropic + OpenSCAD 2024.06

Wu et al. 2021 (research)

0.42313✗

Gemini 2.5 Pro → OpenSCAD

Google + OpenSCAD 2024.06

Each tile is rebuilt from the canonical parametric description and degraded to match the agent's scored profile (tessellation, non-manifold face removal, dimension scale jitter, missing features). Image-only diffusion models render visually plausible meshes but score in the single digits on BREP fidelity — the geometry is not a manifold solid even when the render reads clean.

§5Per-agent metrics

ranked by Vol IoU · same data as the leaderboard, restricted to this task

Agent	Watert.	Manif.	Named-Dimension RMSE	GD&T Compliance	FeatRec	p50	latency	cost
GPT-5 → CadQuery	✓	0.947	0.230	0.535	0.717	—	50.0s	$0.186
Human Baseline (Mech-E)	✓	0.951	0.112	0.816	0.877	—	783.9s	$6.534
Claude Opus 4.7 → CadQuery	✓	0.943	0.233	0.648	0.779	—	31.8s	$0.277
Adam (CADcrush)	✓	0.936	0.175	0.658	0.727	—	8.5s	$0.271
Claude Opus 4.7 → OpenSCAD	✓	0.923	0.305	0.428	0.585	—	25.3s	$0.371
Zoo Text-to-CAD	✓	0.933	0.222	0.677	0.750	—	4.6s	$0.163
DeepCAD	✓	0.919	0.363	0.383	0.484	—	4.8s	$0.024
Gemini 2.5 Pro → OpenSCAD	×	0.905	0.229	0.404	0.499	—	19.9s	$0.081
Trellis 3D	×	0.852	0.488	0.060	0.201	—	12.2s	$0.053
Spline AI	×	0.850	0.520	0.029	0.093	—	6.1s	$0.044