Join AFP testing

AFP’s evaluation work is open for independent review and reproducibility checks. The current public benchmark is a text-only pilot; benchmark scores are not yet published.

Useful ways to contribute

• Review the evaluation protocol and identify validity or scoring problems.
• Reproduce the public benchmark pack from the repository and report discrepancies.
• Propose concrete failure cases with enough context to reproduce them.
• Submit implementation or documentation improvements through GitHub.

How to participate

Use the project’s GitHub repository to open an Issue or Pull Request. Please include the task, environment, model/tool versions when relevant, observed behavior, and enough evidence for another person to inspect the claim.

Open the AFP repository on GitHub

Current boundary

There is currently no separate application form, membership program, guaranteed response time, co-publication promise, or private early-access program. Any future participation process will be documented here before it is offered publicly.