Bring your questions. Build with us.
You do not need a formal academic proposal to make an impactful contribution. An independent reproduction run, an edge-case bug report, or a clearer limitation note moves the entire science forward.
One open source workflow to begin.
Start with a repository README, check the evidence boundaries, and propose small, easily testable changes.
Choose an experimental vector.
Review a project README and its stated limitations. Identify a specific question, boundary condition, or test failure small enough to isolate.
Discuss before implementing.
Open an issue in the target repository to align on scope. Share expected results, hardware constraints, and baseline comparison plans.
Submit reproducible artifacts.
Follow repository contribution rules. Include test suites, random seeds, environment lockfiles, and clear documentation with your pull request.
Small contributions. Real empirical value.
Reproduce an experiment.
Rerun an evaluation protocol on your own hardware. Record hardware discrepancies, seed variance, and whether conclusions hold.
Strengthen the tools.
Refactor runtime tracing, increase test coverage, or benchmark execution replay performance in AgentWatch and PulseStack.
Clarify the protocols.
Document hidden assumptions, clarify setup requirements, or point out unverified claims in our specifications.
Review repository guidance before opening pull requests.
Each project maintains distinct licensing, typing, and testing fixtures. Please consult the dedicated contribution documents below:
Good questions deserve independent reproduction.
Reproduce an experiment, challenge an assumption, or join the discussion with fellow researchers on Discord and GitHub.