HedgeFlow

HedgeFlow is an autonomous, multi-service correlation arbitrage platform designed for trust-aware algorithmic trading. It continuously gathers cross-asset market data, calculates rolling Pearson correlations, and identifies dislocations using z-score deviations from historical correlation patterns. Only statistically meaningful opportunities move forward through strict validation gates, including minimum absolute z-score thresholds, signal confidence checks, and pair-health validation. Qualified signals are then passed to an LLM-based decision layer that returns structured JSON outputs (execute/skip, long-short leg directions, position size, confidence level, and rationale). A deterministic sanitizer then ensures market-neutral leg symmetry, enforces confidence boundaries, and applies size limits, creating fail-closed behavior whenever model outputs are invalid or confidence is too low. Execution relies on EIP-712 typed intents. Each potential trade is checked for freshness and minimum size requirements, signed by the agent wallet, simulated against router risk constraints, and submitted only if the simulation completes successfully. Nonce synchronization along with controlled retry logic helps reduce failures caused by nonce drift or temporary contract rejections. A dedicated settler service tracks transaction receipts, applies a configurable holding window, calculates mark-to-market PnL, and logs results into a local reputation and risk profile. This profile monitors win rate, drawdowns, streak patterns, and cumulative PnL, which then guides circuit-breaker pauses and reputation-based position scaling. It also includes a real-time dashboard that provides clear visibility into agent health, active correlation breaks, trade lifecycle stages, performance metrics, and runtime logs. The outcome is a fully auditable closed-loop system that blends quantitative signal detection, safeguarded AI decision-making, contract-ready execution, and continuous risk-driven feedback.