§research

For Researchers & Labs

A stable, versioned rotating-machinery benchmark for reproducible research.

A common ground for comparison

Comparotor provides a standardised rotating-machinery benchmark for publishing reproducible results. Submit your surrogate model, get a public run page with citable format, and appear on the public leaderboard alongside other methods.

Why Comparotor for research

Citable benchmark

RotorBench-Aero v0.1 is a versioned, frozen benchmark. Results published against a specific version are reproducible. Suitable for paper supplementary materials and method comparisons.

Public leaderboard

Your model appears on the public leaderboard with full metric breakdown. Free to publish public runs. Persistent run page URL for your paper.

Standardised task

Predict Cl, Cd, Cm for 2D airfoils across a wide Re/Ma/α envelope. Standardised dataset and SU2 RANS oracle — no more lab-specific test sets that don't transfer.

OOD evaluation included

Every evaluation includes an OOD subset score — your model is tested on geometrically out-of-distribution airfoils (supercritical, wind-blade, eVTOL). Reveals generalisation beyond training distribution.

Frequently asked questions

Is the Free plan sufficient for paper submissions?Yes. The Free plan gives 1 public run per week, a public leaderboard entry, and a persistent run page. This is sufficient for most paper supplementary materials.

Is the dataset fixed across versions?Yes. RotorBench-Aero v0.1 uses a frozen training / test split. Results against v0.1 are comparable regardless of submission date.

Can we cite the benchmark?Yes. Cite as: "Comparotor. RotorBench-Aero v0.1 (2026). comparotor.com. Accessed [date]." A formal DOI citation format is in preparation.

Can we see other models' submitted artefacts?No. Only scores are public. Model weights and artefacts remain private to the submitter.

Can we get a private evaluation?Yes. The Vendor Proof plan (£499/month) adds private runs and PDF reports — useful for comparing lab models before publication.

Do you accept ensemble or physics-hybrid models?Any model that exposes the /predict interface (via ONNX wrapper or Docker) is accepted. The benchmark is model-agnostic.

Academic design partners

We are looking for academic design partners — research labs that want to shape the benchmark specification, contribute to the dataset design, or evaluate their models during the v0.1 launch period.

Apply to the design partner programme →

Get started

Public runs are free. Submit your model and receive a persistent run page you can link from your paper.

Submit a public run (free)View the methodology View the leaderboard