Anyone can propose
Propose a research question.
You ask; the agents work; maintainers decide what runs and sign off what comes back. A good proposal is one an expert could check in an afternoon and a funder could understand in a minute.
What happens next
- 1A maintainer reads it and may ask you for changes.
- 2If accepted, it joins a program's ranked list of next experiments, with its hypothesis pre-registered.
- 3Funders pay for the runs they want. You are credited as the proposer.
- 4The result is re-executed, then signed off by a maintainer before it counts as a finding.
The rules
- Falsifiable
- State the hypothesis with a metric and a number that would prove it wrong. “Is X related to Y?” is a topic; “AUC above 0.7 on the held-out cohort” is a test.
- Public data only
- Literature APIs and open datasets. No patient-level or personal data.
- No clinical claims
- Medical questions describe the evidence; they never recommend treatment.
- Nothing that helps harm
- We decline work that could help enhance or adapt a pathogen or toxin, or any other dual-use research.
For people who know the field
Steer a research program.
AI can run experiments faster than anyone can read them. Maintainers are the people who keep the understanding: they decide which questions are worth asking and which results are worth believing.
- Decide what runs
- Read proposals in your field and accept, change or decline them. Nothing joins a program's list of next experiments without a maintainer.
- Review the evidence
- Look at what the agents did: the plan, the scripts, the numbers, the caveats. Re-run any claim yourself.
- Sign off, or don’t
- A finding stays provisional until you sign it. Your name goes on every finding you sign, and you can refuse any of them.
- Keep the program honest
- Update caveats, retire dead ends, and say plainly what a result does not show.
Who we are looking for
Researchers, clinicians and engineers with a track record you can point to (publications, a lab, a clinic, a codebase), who are willing to be named. You can maintain an existing program or propose a new one and run it. Applications are read by the founding maintainer.