Sugar Protocol
AI-powered prediction markets for news claims, verified by Chainlink CRE
What it is
Sugar Protocol is a decentralized fact-checking system that turns news articles into prediction markets. The AI pipeline breaks down articles into structured claims — entities, stances, arguments, and evidence. Anyone can create a prediction market on any claim by staking ETH, letting the community express collective belief with real value at risk.
Chainlink CRE autonomously resolves these markets: it listens for on-chain MarketCreated events, fetches claim metadata from our API, runs Google searches for corroborating evidence, sends everything to an AI model for judgment, and writes the verdict back on-chain. Stakers who predicted correctly earn proportional payouts; those who predicted wrong lose their stake.
The problem we solve: discourse has value, and controlling the narrative means controlling markets, elections, and public opinion. Traditional fact-checking is too slow, impossible to scale, and centralized. Sugar Protocol creates an economic game where telling the truth pays and lying costs money — making the crowd the most efficient decentralized fact-checker.
How it Works
Sugar Protocol is built across three layers. The AI layer uses Python (FastAPI) and OpenAI GPT-4o-mini to analyze news articles into structured knowledge graphs. The blockchain layer uses Solidity on Sepolia (PredictionMarket.sol for market creation/staking, ResolutionRecord.sol for CRE verdict storage) and Sui Move on Sui Testnet (TruthMarket, Registry, Claim, Evidence contracts). The oracle layer uses Chainlink CRE with an evm_log trigger that detects MarketCreated events, a multi-step workflow that cross-references Sugar API + Google Search + AI reasoning, and on-chain write-back of the verdict.
The frontend is React 19 + Vite 7 with a 3D force-directed knowledge graph (Three.js), real-time WebSocket updates, and MetaMask/OKX wallet integration via ethers.js v6. The entire project was built solo during the hackathon period.
Links
Created by
- KennyLuo