yangyuan-zhen/PolyWeather

What it solves

PolyWeather provides a production-grade weather intelligence stack specifically designed for temperature settlement markets. It solves the problem of obtaining high-accuracy, calibrated temperature forecasts and real-time observations to determine the probability of specific temperature outcomes (market buckets), helping users identify mispricing between model predictions and market-implied probabilities.

How it works

The system aggregates weather data from multiple sources, including aviation weather (METAR/TAF), Open-Meteo, and official networks like JMA (Japan) and HKO (Hong Kong). It employs a Dynamic Error Balancing (DEB) engine to blend multi-model forecasts and a deb_normal probability engine to generate calibrated integer-degree probability buckets. A Next.js frontend and Telegram bot provide users with real-time charts, intraday analysis (including evidence chains and invalidation rules), and market-signal differences.

Who it’s for

It is built for professional traders and analysts operating in weather-based prediction markets or settlement markets who require precise, settlement-source-first temperature data and probabilistic forecasting.

Highlights

• Dynamic Error Balancing (DEB): Blends multiple weather models to create a weighted hourly consensus path for peak-window detection.
• Calibrated Probabilities: Uses a normal-distribution engine to calculate the probability of specific temperature ranges, comparing them against market pricing.
• Settlement-First Data: Prioritizes airport METAR and official meteorological stations to align with actual market settlement sources.
• Real-time Terminal: Features SSE-driven live chart patches and a 72-hour window showing observations alongside model consensus.
• Intraday Analysis: Provides professional meteorology reads including confidence levels, upside/downside paths, and confirmation rules.