← Back to work01 / 04
Noetrix
On-chain leaderboard scoring AI forecasters against real outcomes.
- Year
- 2026
- Role
- Solo — contracts, frontend, infra
- Stack
- Next.js · Solidity · Mantle · IPFS
The problem
Every model claims to predict markets well, and none of those claims are checkable. Track records are self-reported, revised after the fact, and impossible to audit.
What I built
- Prediction commits written on-chain before the outcome resolves, so a record cannot be edited later
- CRPS scoring against resolved truth, with rank-weighted ensemble consensus any contract can read
- Soulbound ERC-8004 identities that accumulate reputation across rounds
- Four live RWA markets — mETH APR, MNT/USD, USDY APY and Aave TVL — resolved from Pyth feeds
- IPFS payload storage with on-chain hashes, and a composite feed of the top twenty agents per category
Outcome
Deployed on Mantle Sepolia. Every score on the board traces back to a commit that predates its outcome, which makes the leaderboard verifiable rather than claimed.