MECH-014 · Parametric Mechanical Parts · difficulty 3/5

L-bracket with M6 + slotted hole

sha256:1f3a5e90c44b2210…

§1Prompt verbatim

Right-angle L-bracket, leg lengths 60 mm and 40 mm, thickness 5 mm. Through-hole Ø 6.6 mm with 1.5 × 45° chamfer on the long leg, centred 30 mm from the bend. Slotted hole 8 × 16 mm on the short leg, centred 20 mm from the bend, slot major axis parallel to the bend. Position tolerance ±0.1 mm. Output STEP AP242.

§2Ground-truth spec

shells1

genus2

watertighttrue

manifoldtrue

acceptance ε±0.1 mm

featuresbend_R5, thru_hole_M6_clearance, chamfer_1.5x45, slot_8x16

§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)

OpenAI + CadQuery 2.4

0.62510✗

Claude Opus 4.7 → CadQuery

Anthropic + CadQuery 2.4

Claude Opus 4.7 → OpenSCAD

Anthropic + OpenSCAD 2024.06

0.4890✗

DeepCAD

Wu et al. 2021 (research)

0.40914✗

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	P@1	p50	latency	cost
Human Baseline (Mech-E)	✓	0.958	0.078	0.800	0.885	1.000	—	675.9s	$7.058
Zoo Text-to-CAD	✓	0.939	0.140	0.637	0.709	0.000	—	4.9s	$0.170
GPT-5 → CadQuery	✓	0.938	0.201	0.584	0.694	0.000	—	38.7s	$0.235
Claude Opus 4.7 → CadQuery	✓	0.949	0.220	0.611	0.776	0.000	—	35.3s	$0.323
Adam (CADcrush)	✓	0.939	0.252	0.707	0.709	0.000	—	6.8s	$0.256
Claude Opus 4.7 → OpenSCAD	✓	0.926	0.212	0.476	0.527	0.000	—	37.5s	$0.297
DeepCAD	✓	0.917	0.282	0.356	0.477	0.000	—	5.7s	$0.016
Gemini 2.5 Pro → OpenSCAD	×	0.895	0.254	0.425	0.480	0.000	—	32.7s	$0.082
Trellis 3D	×	0.861	0.462	0.060	0.193	0.000	—	15.6s	$0.058
Spline AI	×	0.850	0.579	0.027	0.095	0.000	—	9.5s	$0.035