An open, agent-agnostic harness playing ARC-AGI-3. Stock coding agents, a prediction-graded ledger, and a fenced verifier — every score below links to a scorecard verified on ARC’s servers. Source (MIT-0).
| game | levels | status | card | actions | human median | scorecard |
|---|
Public demo set (25 games). Human median = upper-median first-time human, total actions per game. Card = ARC’s server-side game score on a closed scorecard. A parked game is not abandoned — its ledger, manual, and refuted hypotheses wait for the next attempt.