REVENG-009 · Reverse Engineering · difficulty 5/5

Three-view ortho → housing with cores

sha256:60189cae3b771acc…

§1Prompt verbatim

Reproduce the 80 × 60 × 40 mm housing from the supplied multi-view drawing including all M4 tapped holes, draft, and ribs. Drawing follows ASME Y14.5-2018 third-angle convention.

§2Ground-truth spec

shells1

watertighttrue

manifoldtrue

acceptance ε±0.1 mm

featuresthread_M4_x6, rib_x4, draft_1deg

§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)

n=4 senior engineers

0.60310✗

GPT-5 → CadQuery

OpenAI + CadQuery 2.4

Claude Opus 4.7 → CadQuery

Anthropic + CadQuery 2.4

0.51814✗

Gemini 2.5 Pro → OpenSCAD

Google + OpenSCAD 2024.06

Claude Opus 4.7 → OpenSCAD

Anthropic + OpenSCAD 2024.06

Wu et al. 2021 (research)

0.09948✗

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	Vol IoU	Watert.	Manif.	Named-Dimension RMSE	FeatRec	p50	latency	cost
Human Baseline (Mech-E)	0.603	✓	0.946	0.073	0.932	—	556.1s	$6.762
GPT-5 → CadQuery	0.550	✓	0.938	0.226	0.704	—	38.1s	$0.205
Trellis 3D	0.528	✓	0.925	0.526	0.208	—	14.3s	$0.048
Claude Opus 4.7 → CadQuery	0.518	✓	0.925	0.189	0.767	—	26.8s	$0.305
Gemini 2.5 Pro → OpenSCAD	0.449	✓	0.921	0.248	0.487	—	25.6s	$0.094
Zoo Text-to-CAD	0.377	×	0.910	0.182	0.769	—	6.3s	$0.209
Claude Opus 4.7 → OpenSCAD	0.314	×	0.898	0.210	0.582	—	39.0s	$0.360
Adam (CADcrush)	0.282	×	0.895	0.254	0.705	—	9.0s	$0.231
Spline AI	0.204	×	0.881	0.521	0.086	—	7.8s	$0.032
DeepCAD	0.099	×	0.866	0.303	0.446	—	4.3s	$0.024