PARAM-006 · Constraint Solving & Editability · difficulty 3/5

Editable flange (bolt circle param sweep)

sha256:0ce7b1d445aa9088…

§1Prompt verbatim

Flange: hub Ø 30 × 20 mm, plate Ø 100 × 8 mm, six Ø 7 mm bolt holes on PCD 'D'. Build it once at D = 80 mm, then expose 'D' as a parameter. We will edit D to 70, 75, 85, 90 mm and re-evaluate.

§2Ground-truth spec

shells1

watertighttrue

manifoldtrue

acceptance ε±0.05 mm

parametric edits

D : 80 → 70 (ΔV expected 0 mm³)

D : 70 → 75 (ΔV expected 0 mm³)

D : 75 → 85 (ΔV expected 0 mm³)

D : 85 → 90 (ΔV expected 0 mm³)

§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✓

Adam (CADcrush)

CADcrush

0.8447✓

Claude Opus 4.7 → CadQuery

Anthropic + CadQuery 2.4

0.8358✓

Human Baseline (Mech-E)

n=4 senior engineers

0.83310✓

GPT-5 → CadQuery

OpenAI + CadQuery 2.4

Claude Opus 4.7 → OpenSCAD

Anthropic + OpenSCAD 2024.06

0.5550✗

Gemini 2.5 Pro → OpenSCAD

Google + OpenSCAD 2024.06

0.4970✗

DeepCAD

Wu et al. 2021 (research)

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.	ParamEdit	ConSolve	Parametric Range Integrity	P@1	p50	latency	cost
Adam (CADcrush)	✓	0.974	0.691	0.727	0.701	1.000	—	9.6s	$0.318
Claude Opus 4.7 → CadQuery	✓	0.973	0.790	0.838	0.749	1.000	—	33.7s	$0.358
Human Baseline (Mech-E)	✓	0.975	0.799	0.908	0.818	1.000	—	751.2s	$5.386
GPT-5 → CadQuery	✓	0.950	0.695	0.791	0.671	0.000	—	53.1s	$0.185
Zoo Text-to-CAD	✓	0.939	0.731	0.734	0.631	0.000	—	5.6s	$0.209
Claude Opus 4.7 → OpenSCAD	✓	0.936	0.588	0.629	0.542	0.000	—	29.7s	$0.352
Gemini 2.5 Pro → OpenSCAD	✓	0.929	0.568	0.602	0.545	0.000	—	27.0s	$0.107
DeepCAD	×	0.872	0.280	0.317	0.215	0.000	—	4.1s	$0.023
Trellis 3D	×	0.850	0.057	0.056	0.036	0.000	—	9.6s	$0.041
Spline AI	×	0.850	0.044	0.050	0.020	0.000	—	6.2s	$0.033