What is polymarket-weather-bot?

WeatherBet is a TypeScript bot that trades weather related prediction markets on Polymarket, specifically the daily markets tied to airport temperature readings for cities like New York, Chicago, and Dallas. The core idea is to compare professional weather forecast models against the prices the market is offering, and only trade when the forecast suggests the market price is meaningfully off. The bot pulls forecast data from multiple sources, including the ECMWF and HRRR and GFS models through the Open-Meteo service, along with live airport weather readings from Aviation Weather. It compares the probability these models imply for each temperature bucket against what Polymarket is asking, and only enters a trade when the calculated edge clears a set threshold. Position size is set using a method called fractional Kelly sizing, which scales the bet to the size of the edge rather than betting the same amount every time. The bot also keeps a local record of resolved markets and uses that history to adjust its own confidence over time. Before risking real money, the bot can run in paper trading mode, simulating trades against Polymarket's public prices with a fake balance and no wallet required. Switching to live trading means providing a private key and a Polymarket proxy wallet address in a configuration file, after which the bot places real orders on Polymarket's order book. Risk controls include a stop loss around 20 percent, a rule that trails the stop to breakeven once a position gains 20 percent, and a filter that skips a trade if the bid ask spread is too wide. Everything is controlled through a command line interface with commands to run the main scan loop, check current balance and open positions, or generate a report of past resolved markets. Setup requires Node.js and a clone of the repository, followed by copying an example environment file and installing dependencies with npm. The README describes the software as intended for education and research, states that the person running it is responsible for every trade, and warns that weather markets can move sharply if a forecast model is wrong or a station's data is revised late. It is released under the MIT license.