The Dataset Viewer has been disabled on this dataset.

MARB — Mechanical Assembly Readiness Benchmark

Can AI assemble a real machine?

MARB is a tool-independent, automatically graded benchmark. It measures how correctly AI-assisted CAD places the parts of a real machine: the right position, the right orientation, and the right gap at every interface. The model is given only the authored parts and a set of goal renders — a 3/4 isometric overview plus front, top, and side views, which show how the parts relate in space and reveal components a single view would hide. It receives no build steps. Any tool or agent can be tested, including Autodesk Fusion, CadQuery, or a custom agent.

MARB is developed by Sunnyday Technologies. It runs on the open-source CADCLAW verification engine. Source of truth: github.com/sunnyday-technologies/MARB. Project home: marb.cadclaw.io. Live board: SunnydayTech/marb-leaderboard.

This dataset is the public benchmark input. It contains the blind kits, the goal renders, the frozen task brief, and the scoring spec. It does not contain the answer key. The answer key is a separate, access-gated dataset (gated to prevent training-data contamination, not for secrecy): SunnydayTech/marb-m3-crete-answer-key.

What it grades (v0.9)

MARB grades three positional metrics against the answer key, under a fixed, tight standard:

  • GAP is the error between the actual and intended interface gap. The intended gap is about 0 mm where parts bolt together and about 1 to 2 mm where parts move. GAP is reported as a median in millimeters. It is the primary functional score.
  • ORIENT is the share of orientation-gradeable (asymmetric) parts placed in the correct rotation, reported as a percent aligned. Rotationally symmetric parts are skipped.
  • POS is the position error of each part versus the answer key after a best-fit rigid alignment. It is reported as a median in millimeters, both absolute and neighbor-relative.

Buildability stays on as a secondary gate.

What is in this dataset

  • kits/ — versioned blind kits (one zip per version). A kit holds the authored parts, the goal renders (a 3/4 isometric overview plus front, top, and side views), and the task brief. No answer key. Versions are tracked in kits/KIT_VERSIONS.md.
  • prompts/ — the frozen task brief and the per-backend driver stubs.
  • spec/MARB_SCORING.md — the canonical, versioned scoring method.
  • benchmark.yaml — gate weights for the secondary buildability score.

The grader code lives on GitHub, depends on the cadclaw PyPI package, and is kept there as the single source of truth.

Run a model against it

  1. Give the model a blind kit from kits/. It holds the authored parts, the goal image, and the task brief, with no answer key.
  2. Run the model in a sealed, memoryless context so the answer key cannot leak. Use a neutral folder with cross-session memory turned off.
  3. Hand the exported STEP file back and grade it.

To grade, install the engine and request the gated answer key:

pip install "cadclaw>=0.9.0"
# Download the gated answer key (see the answer-key dataset card for access).
python grader/marb_pose_metric.py   --ref m3_reference_round1.step --run your_export.step
python grader/marb_gap_metric.py    --ref m3_reference_round1.step --run your_export.step
python grader/marb_orient_metric.py --ref m3_reference_round1.step --run your_export.step

Grader source: github.com/sunnyday-technologies/MARB/tree/main/grader.

Versioning and comparability

Results are comparable only within a single kit cohort and a fixed scoring version. Each run records its kit version, its model and tool versions, and the client environment. Do not pool runs across kit versions without noting it.

License

The MARB code, brief, and spec are MIT. Copyright (c) 2026 Sunnyday Technologies. The CAD parts bundled in the kits are licensed separately: OpenBuilds-derived parts are CC BY-SA 4.0, and the two Sunnyday-authored parts are MIT. Full notice: kits/LICENSE.md. V-Slot, C-Beam, and V-Wheel are registered trademarks of OpenBuilds LLC.

Citation

If you use MARB in published research or derivative work, please cite it. See the CITATION.cff in the source repository.

Contact: info@sunn3d.com

Downloads last month
18

Space using SunnydayTech/marb-m3-crete 1