Prediction-Market Arbitrage Engine

The premise: the same real-world event is often priced differently on Polymarket and Kalshi, and the gap should be capturable. Modeling it honestly meant the execution-cost side mattered more than the signal; fees, slippage, and settlement timing routinely ate spreads that looked like free money on paper. The expected-value layer had to account for resolution risk too: two venues can word the 'same' contract just differently enough that they don't actually resolve together. I kept this strictly a research prototype, so there was never live capital or realized P&L; the value was the modeling discipline, not a deployment.