The weather model that surfaces where forecasts and markets diverge

We take NOAA's gridded forecast as the starting point, then apply a per-city adaptive correction based on how NOAA has systematically erred for that specific station recently. The metric that matters for Kalshi is bucket accuracy — how often our predicted bucket matches the actual CLI bucket. Our model hits the right bucket ~74% of the time across the last 30 days. Forecast accuracy in early morning is lower (more uncertainty); accuracy tightens as more observations arrive through the day. Full per-decile calibration is published at api.sardine-weather.com/calibration.

It's wrong sometimes — every forecast is. We publish full accuracy in the Accuracy tab, per city and per time-to-peak band. You see misses the same day they happen, no curation. If a city has a track record you don't trust, trade a different one.

No. Sardine is a data tool — the Kalshi bucket prices are shown for reference so you can see where market pricing disagrees with the model. If you want to trade, each bucket row links directly to the Kalshi market. We don't route orders, hold funds, or take a cut.

Real-time model output (no 10-minute delay), edge and fair-value per bucket for every city, confidence tiers that flag when the model is most reliable, per-city historical performance, and direct tap-to-open Kalshi links. Free tier shows NOAA + current obs + one fully unlocked demo city so you can see what the experience is like before paying.

Mostly, yes — the pricing reflects that. If you're someone who tracks the peak daily temperature (utility operators, logistics, HVAC ops, anyone long-gamma on heat), the near-peak forecast accuracy (~74% bucket hit) is useful even if you never touch a Kalshi market. But trader is the core user.