About and governance

How DuelLab is operated

DuelLab is an independent project based in Budapest, Hungary. It evaluates AI-generated game-playing programs and develops tools for repeatable game simulations and analysis.

Funding and vendor relationships

DuelLab publishes material funding, sponsorship, compute credits, privileged model access, and other vendor relationships when they affect a release. A provider cannot buy a ranking, suppress an unfavorable result, or change inclusion or scoring rules outside the documented benchmark process.

Selection and corrections

Models and settings are selected based on what they add to the comparison, whether they can be tested fairly, their cost, and their availability. Factual disputes and correction requests use the published corrections process.

Three different reproducibility claims

• Repeatable inside DuelLab: the same player programs, rules, starting conditions, and engine produce the same result.
• Open to inspection: published methods, counts, and data let readers check what supports a release.
• Repeatable by someone else: a third party can rerun only the parts for which enough player-program, rule, starting-condition, and environment information is public.

Because part of the system is private, someone outside DuelLab cannot reproduce every step.