Kalshi weather bot: what it does, how it prices, where it stops

What it does

A FastAPI backend with SQLite and APScheduler runs the scans; a React dashboard shows indicators, weather panels, signals, trades, an equity curve and a calibration panel. Every data source is free: Coinbase, Kraken and Binance candles, Open-Meteo ensemble forecasts, NWS observations, Polymarket Gamma and Kalshi.

Kalshi weather markets are auto-discovered from the KXHIGHNY, KXHIGHCHI, KXHIGHMIA, KXHIGHLAX and KXHIGHDEN series. Adding a city means editing WEATHER_CITIES and CITY_CONFIG.

• Weather scan every 300 seconds, BTC scan every 60 seconds
• Edge thresholds: 8% weather, 2% BTC
• Fractional Kelly at 0.15, capped at 5% of bankroll and $100 (weather) or $75 (BTC) per trade
• Daily loss circuit breaker $300, max 20 pending trades, $10,000 starting paper bankroll

Strategy

Weather: count the ensemble members above or below the market threshold, and that fraction is the model probability. 28 of 31 members above 70F is a 90% estimate. Confidence is ensemble agreement. Settlement is checked against NWS observed temperatures.

BTC: fetch 60 one-minute candles, compute RSI(14), momentum at 1, 5 and 15 minutes, VWAP deviation, SMA crossover and market skew, require 2 of 4 indicators to agree, and map the composite to an UP probability between 0.35 and 0.65.

Setup and requirements

Python 3.10 or newer for the backend (uvicorn backend.api.main:app), Node for the frontend (npm run dev). A Kalshi API key ID and RSA private key path are optional and only needed to fetch Kalshi markets; no Polymarket credentials are documented.

There is a hosted dashboard at polymarket-kalshi-weather-bot.vercel.app, which returned 200 on the check date.

Activity and maintenance

740 stars and 185 forks as of 2026-09-08. Created 2026-02-10, last push 2026-03-02, 70 open issues. The GitHub API reports no license file even though the README carries an MIT badge.

Six months without a push and 70 open issues means treat it as unmaintained. Kalshi series tickers and Polymarket market shapes change; expect to fix the fetchers.

Where it falls short

The README documents simulate-trade and a paper bankroll but no live order path, and its disclaimer says it does not place real trades. The $1.8k figure in the repo description is not reconciled anywhere in the README.

Five US cities only. Kalshi's temperature markets settle on specific stations, and the ensemble is a global model, so station bias is on you.

Who should use it

Someone who wants a working example of ensemble-to-probability pricing for weather contracts and a calibration loop, and is prepared to write the execution layer and maintain the fetchers themselves.

Not for anyone expecting a live bot out of the box.