FUNC-007 · Functional Intent · FEA-Gated · difficulty 5/5

Heat-sink fin array for 25 W TO-220

sha256:a012b3e44c510fd0…

§1Prompt verbatim

Design a fin-array heat sink in 6063-T5 aluminium that holds a TO-220 device (mounting hole pattern given) below 95 °C junction at 25 W in still air at 25 °C ambient. Footprint ≤ 60 × 60 mm, height ≤ 35 mm. Output STEP.

§2Ground-truth spec

shells1

watertighttrue

manifoldtrue

acceptance ε±0.1 mm

featuresTO220_mount, fin_array

§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✓

Human Baseline (Mech-E)

Claude Opus 4.7 → CadQuery

Anthropic + CadQuery 2.4

0.47813✗

GPT-5 → CadQuery

OpenAI + CadQuery 2.4

Claude Opus 4.7 → OpenSCAD

Anthropic + OpenSCAD 2024.06

0.1760✗

Gemini 2.5 Pro → OpenSCAD

Google + OpenSCAD 2024.06

0.1460✗

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.	FeatRec	Min-Wall Compliance	FEA-Yield Pass	p50	latency	cost
Human Baseline (Mech-E)	✓	0.931	0.846	0.889	0.753	—	755.3s	$5.617
Zoo Text-to-CAD	✓	0.918	0.807	0.691	0.479	—	6.7s	$0.194
Claude Opus 4.7 → CadQuery	✓	0.922	0.755	0.612	0.528	—	31.1s	$0.355
GPT-5 → CadQuery	✓	0.916	0.648	0.637	0.524	—	41.0s	$0.247
Adam (CADcrush)	×	0.900	0.678	0.653	0.404	—	8.6s	$0.250
Claude Opus 4.7 → OpenSCAD	×	0.877	0.561	0.546	0.387	—	23.0s	$0.318
Gemini 2.5 Pro → OpenSCAD	×	0.872	0.520	0.564	0.387	—	28.9s	$0.098
DeepCAD	×	0.866	0.492	0.364	0.191	—	6.0s	$0.016
Trellis 3D	×	0.850	0.205	0.206	0.085	—	14.9s	$0.059
Spline AI	×	0.850	0.099	0.094	0.046	—	6.1s	$0.035