WeatherQuant

~143 ensemble members, not 3 deterministic runs

We pull the GEFS (31), ECMWF ENS (51), ICON-EPS (40), and GEM (21) ensemble systems from Open-Meteo — roughly 143 possible futures per city. We don't average them into one number; we use the whole distribution.

Probability per bracket, directly

For each Kalshi bracket, we count how many ensemble members land there. If 42 of 143 members say the high will be 72–73°F, that bracket has a 29% probability. Simple, honest math — no hand-tuned weights.

EV = our probability ÷ market price − 1

If our probability is 29% and Kalshi is pricing the bracket at 20¢, EV = +45%. Positive EV means the bracket is underpriced vs. our model. The opposite is also true — we'll show you brackets the market is overvaluing.

Bias correction + live observation override

Every time Kalshi settles a market, we record the error between our ensemble mean and the actual high. That bias gets subtracted from future forecasts for that city. After 1 PM local, if the observed temp is already higher than any member predicted, we clip the distribution upward.

Signal types

This is not financial advice

Ensemble forecasts are not oracles. Brackets can and do settle at $0. Never risk what you can't lose. Track record below is raw — no cherry-picking.