polymarket-trading

This repo implements a sophisticated arbitrage strategy that goes beyond simple sum-of-probabilities checking. It treats the arbitrage execution as a convex optimization problem: given multiple outcome markets with varying liquidity and order book depth, the goal is to allocate a fixed capital across the arbitrage to maximize expected return under slippage constraints. The Frank-Wolfe algorithm iteratively finds the optimal trade vector, considering each market's order book shape and cross-market correlations. Implementation is TypeScript with numeric optimization libraries (likely simple stats or custom implementation). Key components: order book fetcher from CLOB API, convex problem formulation module, Frank-Wolfe solver, and execution engine. At 1.4MB it includes a paper. The mathematical approach is sound and novel for Polymarket bots — most competitors use simple threshold checks. Weaknesses: no tests visible, limited documentation in English, and convex optimization may be overkill for typical arb spreads (which are usually <1%). The solver also requires careful tuning of convergence parameters. This is an academic-grade tool best suited for quant researchers and mathematicians. Real-world performance depends on execution speed more than optimization sophistication. No malware risk.