zeus
github.com/fitz-s/zeus
Weather-derivatives trading engine for Polymarket with multi-model forecast fusion, settlement-aware probabilities, walk-forward calibration, and risk-sized execution.
Bets on temperature outcomes using an ensemble of forecast models (METAR, Open-Meteo) fused into calibrated bin probabilities, with settlement-aware adjustments and risk-sized fail-closed execution.
Zeus is a serious weather-derivatives trading engine that stands out for its engineering depth. Unlike simple METAR-vs-forecast bots, it employs multi-model forecast fusion to build a probabilistic view of temperature outcomes, then applies settlement-aware bin probabilities to align positions with how Polymarket actually resolves these markets. The walk-forward calibration loop is a notable feature, ensuring the model's probability estimates are continuously validated and corrected against live market outcomes. Risk-sized execution with a fail-closed design indicates careful thought about capital preservation in a market with fat tails. The codebase appears well-structured for a 75MB repository, suggesting it may include historical data or model artifacts alongside the core engine. For traders interested in weather markets, this is among the most technically complete open-source examples, offering a blueprint for a full systematic trading pipeline. The main caveat is that running a 24/7 live bot requires robust infrastructure and careful monitoring. The primary risk is model overfitting to historical weather data, but the walk-forward approach mitigates this. It is a strong candidate for study and adaptation, especially for those looking to move beyond manual trading to a calibrated, automated system.