risedownlabs/polymarket-weather-bot

This repository is another implementation of WeatherBet, a TypeScript bot that automates trading on Polymarket's daily temperature prediction markets. Like similar weather trading bots, it works by comparing forecasts from multiple weather models against the prices Polymarket is currently offering, and only trades when it calculates a meaningful edge. The bot pulls forecasts from the ECMWF and HRRR and GFS models by way of the Open-Meteo service, along with live airport weather readings called METAR from Aviation Weather. Markets resolve against specific airport station codes, such as KLGA for LaGuardia, rather than a general city location, so the bot maps each market to the correct station. Before placing a trade it checks that the expected value clears a set threshold and that the bid ask spread is not too wide, then sizes the position using fractional Kelly sizing, which scales the bet to the size of the calculated edge. A stop loss around 20 percent and a rule that moves the stop to breakeven after a 20 percent gain are both built in to manage risk once a position is open. The bot can run in two modes. Paper mode simulates trades using Polymarket's public prices and a fake balance, requiring no wallet at all, and is described as the way to start. Live mode requires a Polygon wallet private key and a Polymarket proxy wallet address supplied through a configuration file, after which the bot places real orders on Polymarket's order book. Resolved markets are saved locally and can be reviewed with a report command that shows win rate and how well past forecasts calibrated against outcomes, which the README suggests checking before switching to live mode. Setup requires Node.js version 20.10 or newer. After cloning the repository, a person installs dependencies with npm and copies an example environment file to configure their settings. Everything after that runs through a small set of command line commands to start the scan loop, check status, or pull a report. The README lists clear limitations: a wrong forecast model, a relocated weather station, or a late correction to a METAR reading can all cause a market to resolve differently than expected, and the fills and slippage seen in live trading will differ from the simulated prices used in paper mode. The project is released under the MIT license.