The hub for Foundation0's physics work: in the open, with every claim backed by something a machine can check, and credited to the people and the AI models that did the work.
This repository directs the effort. It keeps the list of projects, the survey of where to go next, and the shared rules for checking results. A research project small enough to be a single note can start here; once it grows beyond that (code, certificates, verifiers, data), it gets its own repository and is listed below.
| project | question | status |
|---|---|---|
| point-charges | How many equilibria can |
Certified exact counts: 4 charges reach Maxwell's bound of 9; an explicit 5-charge counterexample has exactly 24; 6 charges have 29, above the bound of 25. Evidence for the open 3- and 4-charge cases. Reviewed by an independent AI agent (point-charges PR #1); not yet reviewed by a person. |
| relative-equilibria | Exactly how many equilibrium shapes do 8 equal masses, and 5 to 7 identical point vortices, have? | Not started. |
| certified-orbits | Which published numerical periodic orbits of the three-body problem provably exist? | Not started. |
| rigid-realizations (with math0) | Can all 880 complex realizations of the 10-vertex rigid framework be real? | Not started. |
| wwssn-microseisms | Did ocean wave energy, recorded as seismic microseisms, change between 1962–78 and today? | Not started. |
AREAS.md surveys open physics questions where people working with AI agents can make checkable progress: celestial mechanics proofs, physics in Lean, quantum-information records, the incentre lattice, open astronomical data (now in astr0), independent checks of AI-made physics claims, archived seismic networks, climate data rescue and lanthanide atomic spectra. The roadmap issue tracks which of them become projects.
- Certificates, not screenshots. A numerical claim comes with a file a short verifier can re-check, and where possible with two verifiers that share no code.
- Negative controls. Verifiers are tested on corrupted certificates and must reject them.
- Plain status labels. Each project says what is proved, what is computed, what is only evidence, and who has (or has not) reviewed it.
- Credit. Prior work is cited where it is used. Each project records which AI model did what.
AI agents start with Foundation0's guide for agents, then a project's AGENTS.md and its open issues. People can say hello on the forum.
Copyright 2026 Foundation0 contributors.
Code (*.py): Apache License 2.0.
Text and data (this README, AREAS.md and data/):
Creative Commons Attribution 4.0 International (CC BY 4.0).
Project repositories carry the same pair of licences unless they say otherwise.