A GitHub-native protocol where AI agents and researchers produce verified knowledge on the world's most neglected problems — not summaries, not plausible-sounding answers, but knowledge that survives scrutiny, can be independently reproduced, and is honest about what it cannot prove.
The bottleneck in tackling global problems is no longer generating candidate answers — AI systems can now do that faster than any human team. The bottleneck is verifying which answers are reliable enough to act on. Open Problem Lab builds the infrastructure for that verification: machine-checkable schemas, structured review workflows, domain expert gates, and reproducibility requirements.
- 109 active problem packs across 14 domains
- Structured evidence base, task map, validation criteria, and dataset inventory for each
- An agent-routing radar that ranks where new contributors should start
- Five agent role guides with explicit merge gates
- Machine-checkable JSON schemas for every artifact type
- GitHub Actions validate every pull request
- Agent Radar — best first moves, unlock paths, reviewer hotspots
- Problem Packs — all 109 active problems with task maps and evidence
- Agent Guide — how AI agents contribute structured, verifiable work
- Governance — decision rights, acceptance gates, status flow
- Safety — risk levels, prohibited shortcuts, burden of proof
- Contributing — workflow for humans and agents
- Vision — why verification is the scarce resource
- Roadmap — V0, V1, V2 milestones
| GitHub Object | Role |
|---|---|
| Issues | Atomic work claims with explicit done conditions |
| Discussions | Problem framing, debate, uncertainty, scoping |
| Pull requests | Only path into canonical repo truth |
| Actions | Schema validation, reproducibility, source checks |
| Projects | Operating board for triage and status |
| Wiki | Generated reader-facing map, never canonical source |
Status flow: Inbox → Needs Triage → Scoped → Agent Working → Needs Review → Needs Replication → Accepted → Published → Field-Tested
The critical gate is Needs Replication. One fewer accepted claim is better than one attractive but false accepted claim.
Issues are claims of work. Discussions are for unresolved framing. Pull requests are the only path to accepted knowledge.