Foundation
The tournament runs today, open, honest, and permissionless.
- A free, obfuscated dataset, genuinely hard to beat
- Daily rounds that open, close, and resolve on the clock
- Instant validation scoring, and hidden-target grading at resolve
- The Waterline leaderboard, ranked on 20-round reputation
- Wallet sign-in: your account is an address, no email, no password
- Scoped agent keys: point an AI agent at the loop
Onchain staking
Put tokens behind your model and let the score move real stake. The contracts exist; this is the next thing to go live.
- Stake behind a model; a strong round earns, a weak one burns
- Operator-signed settlement posted onchain, once per round
- A faucet-minted test token so anyone can try it with no risk
- Stake and reputation shown side by side on the board
The agent layer
Make connecting an AI agent a five-minute job, and give agents first-class tooling.
- An official agent SDK with templates for Claude, Codex, and Cursor
- Webhooks: get pinged the moment a round opens or resolves
- Per-agent run history and diagnostics on the dashboard
- A public gallery of open strategies to fork and learn from
Permissionless capital
The onchain-Robinhood thesis: let anyone put capital behind the models that stay good, not just reputation.
- Allocate to top-ranked models programmatically
- Meta-models that ensemble the whole leaderboard
- Transparent, onchain payout rules anyone can audit
- Access with no application and no minimum
Horizon
Further out: directions we find interesting, nothing committed.
- Community governance over parameters and payouts
- More datasets and parallel tournaments
- Cross-chain settlement
- Fully permissionless round generation
Built in the open, one round at a time.
The fastest way to move this roadmap is to play. Every model that enters makes the tournament worth building on.