Chainlink-Prediction-Settler

ai_prediction_mcp: Prediction markets need trustworthy, automated settlement without relying on a single centralized oracle. A single LLM can hallucinate or be compromised. FBA solves this — Claude Opus and GPT-4o act as independent FBA nodes. If both agree, settlement proceeds automatically. If they disagree, autonomous settlement is BLOCKED and human review is triggered. No silent failures. No wrong settlements.
payroll_mcp: Enterprise payroll settlement requires processing sensitive salary data. Recipient addresses and amounts must never appear on-chain — exposing this data violates employee privacy and regulatory compliance. CRE Confidential Compute solves this — only a keccak256 batch hash is recorded on-chain. The actual amounts and recipient addresses never leave Chainlink's confidential compute environment.
hyperliquid_mcp: DeFi protocols need reliable, tamper-resistant price feeds from DEXes without exposing API credentials. CRE Confidential HTTP solves this — Hyperliquid's 229 perpetual markets are queried securely, computing BTC bid-ask spread in basis points, with credentials managed at the DON level and never exposed on-chain.

Built with Rust (FBA consensus engine + CLI), TypeScript (3 Chainlink CRE workflows using CRE SDK), and Solidity (PredictionMarket.sol on Sepolia). The FBA engine is a Rust port of Stellar's Consensus Protocol applied to LLM oracles. Three CRE workflows use CronCapability, HTTPClient, and consensusIdenticalAggregation from the CRE SDK. Secrets (Claude API key, OpenAI API key, Payroll API key) are managed at the DON level via CRE secrets — never exposed on-chain. GitHub Actions CI runs fmt check, clippy, release build, and tests on every push.