The scanner queries Polymarket's Gamma API using the weather tag filter, discovering 200-1200+ active markets across global cities. Each market has multiple temperature outcomes with prices set by traders.

Each market title gets parsed to extract the city (mapped to lat/lon), date, temperature unit (°C/°F), and outcome options. Handles formats like ranges ("30-31°F"), exact values ("14°C"), and tail outcomes ("15°C or higher").

Four parallel API calls to independent weather models, each returning hourly temperature arrays for the target date.

Models are blended using ECMWF-weighted consensus with outlier detection. The edge engine uses Dynamic σ, Bayesian probability, Normal CDF, and Kelly Criterion to calculate the optimal trade — comparing our forecast probability against the market's implied probability.

Claude operates as a Senior Meteorologist with 15 years of experience. Each analysis follows a 5-step process: individual model examination with known biases, synoptic pattern assessment, microclimate factors (urban heat island, coastal effects), Weather Underground station awareness, and market pricing comparison. Running on Claude Sonnet 4.6 (latest).

Trades execute via Polymarket's CLOB API with FOK orders and GTC fallback. WebSocket streaming monitors all positions in real-time. The exit engine manages stop-losses, trailing stops, and profit targets. Positions and balances sync with Polymarket every 2 minutes.