Contribute
There are three ways to help: run and submit a result, contribute a pack/profile/ adapter, or flag something that looks off. All three matter equally — a benchmark is only as trustworthy as the results and the scrutiny behind it.
1. Submit a benchmark result
The easiest contribution: run the open runner on your own hardware and submit the result. Every submission is re-verified before it's accepted — see the how-to guide. No account or sign-up needed — trust comes from a cryptographic re-check, not identity. Not sure what's missing? See most wanted benchmarks — the public list of gaps other people have asked for.
2. Contribute code: adapters, metrics, packs, profiles
New runtime adapters, metric plugins, hardware probes, normalization rulesets, and benchmark packs are all reviewed, in-tree code merged via pull request — the runner never executes anything supplied at run time. This is a deliberate security boundary, not a formality:
- Security — no arbitrary code execution path introduced.
- Neutrality — no hardcoded assumption favoring the contributor's own hardware, runtime, or model.
- Correctness — matches its documented definition, has tests.
- Reproducibility — never silently changes how an existing official profile scores; a scoring change is always a new profile version.
Adding a new language means contributing a profile plus a normalization ruleset — the core stays language-agnostic by design, so this never requires changing the platform itself. Contributing a profile or pack has its own detailed walkthrough — including an in-browser builder that generates a ready-to-submit file — see Create a profile or pack.
3. Report a neutrality concern
If a result, profile, ranking, or piece of documentation looks like it favours a vendor, runtime, or sponsor, say so. These reports are treated as high priority — neutrality is the product.