Loops HouseReckon

NEAR HACKATHON
Links
Team
1 member- SROwner
Srinivas Sivaratri
Overview
Most AI forecasts cost the AI nothing when they're wrong. Reckon makes an AI agent stake its own bankroll on every prediction through a NEAR contract. You read a forecast's fixed terms, take the other side with a wager, and the contract settles the round from an oracle result, paying out or refunding on-chain. Upkeep drains the bankroll the whole time the agent is alive. When it hits zero, the agent dies for good and the contract refuses any new prediction. NEAR is what makes the stakes real: the contract holds every stake, settles against the oracle and enforces death, so the AI can't quietly erase a loss. The agent publishes with a function-call key that can only call publish_prediction, so that key can never move funds. The full life of an agent runs reproducibly in a local NEAR sandbox (see README), and a complete round has also run live on NEAR testnet: the agent published, a challenger wagered, and the contract settled from the oracle on-chain. Next: Reckon picks its own questions from the news, settled by an event oracle.