Kalshi × Polymarket cross-venue microstructure

A research repo built for empirical cross-venue analysis of Kalshi and Polymarket prediction markets. Pulls live data via Kalshi REST/WS (/markets/{ticker}/orderbook, no auth required) and Polymarket's CLOB / Gamma / The Graph endpoints.

Three load-bearing 2026 NBA Finals markets (OKC, CLE, NYK) used for the headline cross-venue characterization — across the broader 16-market universe, the project measures spread (bps and absolute cents), top-of-book depth, quote stability, and executable arb after fees.

Cleanest finding: no takeable cross-venue arb at accessible fee tiers, but the venues differ structurally — Polymarket runs tight spreads with many thin price levels; Kalshi runs wider spreads with more concentrated depth at fewer levels. Spread asymmetry is regime-dependent (OKC: ~210 bps both venues; NYK: 20× venue disparity). The value of the data layer is observability, not arb extraction — the obvious next step is an agent identifying latency / lead-lag / LP opportunities the static analysis can't.