Loops HouseReceiptz

NEAR HACKATHON
Links
Team
1 member- NEOwner
Neil Khedekar (0xtitan6)
Overview
AI already fact-checks X at scale, but no one can see how a verdict was reached, or whether the operator re-ran it until it got the answer they wanted. Receiptz makes AI fact-checks auditable.
Reply to any post with "@checkreceiptz is this true?" A triage model extracts the claim. Before any evidence is read, the triage agent's NEAR account registers the run (claim hash, post ID, jury config) in our contract. A router picks a primary record from an allowlist (BLS, BEA, Census, SEC EDGAR, Congress.gov, NEAR/Ethereum RPC, DefiLlama), with Wikipedia and web search as lower-tier context. Three jurors from three labs (Zhipu GLM, Alibaba Qwen, DeepSeek on NEAR AI Cloud) vote independently, a red team attacks the verdict, and a deterministic auditor rejects any citation not in the evidence. The jury agent's account then submits the verdict and the receipt's content hash on-chain, and the bot replies with a link.
NEAR is the referee: the contract accepts one verdict per run, marks late verdicts, and reports runs with no verdict as abandoned, so re-rolling a check leaves a public trail. The lookup site re-hashes every receipt against the on-chain CID: VERIFIED, or TAMPERED if one byte changed.
Live on testnet: receiptz-0xtitan6.testnet · site http://52.17.217.82:3000 · bot @checkreceiptz