BOOL-003 · Boolean Robustness · difficulty 4/5
Coplanar-face union (knife-edge stress)
sha256:9a1b21d83fc04421…
§1Prompt verbatim
Two 20 × 20 × 20 mm cubes placed so their +X / −X faces are exactly coplanar. Union them and fillet the shared edge at R 1 mm. The result must be a single watertight body.
§2Ground-truth spec
shells1
V−E+F2
genus0
watertighttrue
manifoldtrue
acceptance ε±0.05 mm
Stresses ε-tolerance handling — many kernels emit a sliver face here.
§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)
Human Baseline (Mech-E)
n=4 senior engineers
0.8237✓
Adam (CADcrush)
Adam (CADcrush)
CADcrush
0.6668✗
GPT-5 → CadQuery
GPT-5 → CadQuery
OpenAI + CadQuery 2.4
0.6178✗
Gemini 2.5 Pro → OpenSCAD
Gemini 2.5 Pro → OpenSCAD
Google + OpenSCAD 2024.06
0.6140✗
Claude Opus 4.7 → CadQuery
Claude Opus 4.7 → CadQuery
Anthropic + CadQuery 2.4
0.5929✗
Zoo Text-to-CAD
Zoo Text-to-CAD
Zoo (KittyCAD)
0.58513✗
Claude Opus 4.7 → OpenSCAD
Claude Opus 4.7 → OpenSCAD
Anthropic + OpenSCAD 2024.06
0.5430✗
DeepCAD
DeepCAD
Wu et al. 2021 (research)
0.36817✗
Trellis 3D
Trellis 3D
Microsoft Research
0.0490✗
Spline AI
Spline AI
Spline.design
0.0440✗
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. | Euler-Poincaré Compliance | P@1 | p50 | latency | cost |
|---|---|---|---|---|---|---|---|---|
| Human Baseline (Mech-E) | 0.823 | ✓ | 0.970 | ✓ | 1.000 | — | 834.9s | $6.512 |
| Adam (CADcrush) | 0.666 | ✓ | 0.949 | ✓ | 0.000 | — | 11.6s | $0.302 |
| GPT-5 → CadQuery | 0.617 | ✓ | 0.942 | ✓ | 0.000 | — | 44.2s | $0.190 |
| Gemini 2.5 Pro → OpenSCAD | 0.614 | ✓ | 0.947 | ✓ | 0.000 | — | 23.6s | $0.083 |
| Claude Opus 4.7 → CadQuery | 0.592 | ✓ | 0.944 | ✓ | 0.000 | — | 48.9s | $0.303 |
| Zoo Text-to-CAD | 0.585 | ✓ | 0.937 | ✓ | 0.000 | — | 5.9s | $0.183 |
| Claude Opus 4.7 → OpenSCAD | 0.543 | ✓ | 0.933 | ✓ | 0.000 | — | 32.0s | $0.370 |
| DeepCAD | 0.368 | × | 0.902 | × | 0.000 | — | 5.8s | $0.021 |
| Trellis 3D | 0.049 | × | 0.858 | × | 0.000 | — | 14.9s | $0.059 |
| Spline AI | 0.044 | × | 0.856 | × | 0.000 | — | 9.6s | $0.047 |