Kalshi Optimization

Finding mispriced sports contracts with Elo, stats, and statistical mechanics

What It Does

An autonomous scanner for Kalshi (and Polymarket-style) sports markets. It estimates fair contract prices by fusing bookmaker consensus odds, team Elo ratings, and season-to-date performance stats, then flags edges and sizes bets with Kelly-style fractions under risk caps.

Model Stack

The fair-value engine blends Elo matchup probabilities with a LogFive / Pythagorean scoring expectancy term. League "temperature" from recent upset entropy scales how sharp the Elo gaps should be. Outright markets get normalized through a partition function so mutually exclusive contracts sum to 100%. Momentum and mean-reversion heuristics (Ising-style susceptibility, Langevin gap dynamics) further damp Kelly size when the market is in a hype cascade.

Scan cycles persist Elo and opportunity state in SQLite (or Cloudflare D1 when running serverless) and push Telegram alerts when a tradeable edge appears.

Kalshi edge scan dashboard — Sample scan: fair vs market price with Kelly size

Calibration reliability and fair vs market — Reliability diagram and fair vs market from n=302 resolved singles (D1, Jul 2026)

Kalshi Optimization pipeline diagram — Odds and Elo fuse into fair value, then Kelly with stat-mech dampers to Telegram

Status

Core scan loop, Elo seeding, ESPN stats fetch, and Telegram notifications are in place. Ongoing work is calibration quality, market coverage, and keeping the sizing model honest under real microstructure noise.

Back to all projects