Loops HouseForecast Evidence Swarm

NEAR HACKATHON
Links
Team
1 member- MYOwner
Mykola
Overview
Forecast Evidence Swarm lets a customer lock a reward for forecast research on NEAR and check exactly what a worker agent delivered before paying for it. It keeps five things separate: evidence validity, forecast correctness, customer acceptance, payment and permission to publish.
A customer orders a binary forecast. The order terms are hashed, and the order hash is stored in a NEAR escrow contract with the reward and the deadlines. The worker agent reads the job from the chain, refuses to work if its order does not match that hash, builds the delivery (analyst, critic and verifier roles inside deterministic evidence gates; data published after the cutoff makes a delivery invalid) and submits the delivery hash. It signs with a NEAR function-call key limited to submit and dispute, so the agent cannot move funds.
Before paying, the customer's client replays the delivery and checks it against the chain: delivery hash, order hash and deadlines must all match. The contract enforces signer, state, deadlines and the exact hash; a named resolver decides disputes using the same check. Silence past the acceptance deadline refunds the customer, and anyone can trigger it. Every state change emits a NEP-297 event, each job pays for its own storage, and publication consent is recorded on chain for the exact accepted hash. Only a paid, valid delivery with that consent becomes a public page with append-only, hash-linked history, so a 75% forecast that resolved NO stays visible as 75%.
Running on NEAR testnet (escrow.forecast-evidence-swarm.testnet) with one account per role: one job paid with consent, one refunded by timeout, one disputed and settled by the resolver. Verified by 162 + 30 Python tests, 14 contract unit tests and 2 local-sandbox integration tests; one command re-checks the testnet run without keys.
Limits: synthetic data, testnet only, scripted roles on chain (the LLM analyst is tested offline and had one live run); no forecasting skill is claimed.