Methodology
The production scanner uses live market ladders, settlement-station observations and named public forecast models. The simulated Sonde profile lens lives only on the demo scanner.
Where each number comes from
Forecast buckets
Kalshi defines the bucket ladder. Public models do not publish a bucket ladder. Sonde maps each model’s high or low temperature into the Kalshi bucket it lands in, and also derives a fair value by putting uncertainty around that model value.
The model bucket is the cleaner trader read: GFS lands in one bucket, NBM lands in another, and those bucket picks can be tracked over time. The fair value is a pricing aid, not a claim that the model itself emitted a probability ladder.
Uncertainty and fair value
Fair value for a bucket is the normal integral over that bucket, centered on the selected model forecast. Sonde derives the displayed pricing uncertainty from time to the expected high or low and disagreement among the available models. For a selected public model such as GFS or NBM, Sonde widens that base uncertainty because the model supplies a point forecast, not its own probability ladder. The displayed uncertainty is a Sonde pricing assumption. It is not published by the selected model.
For supported next-day GFS and NBM forecasts, Sonde replaces that fallback with measured historical error. The calibration groups forecasts by station, model, high or low, lead time, and season. Sonde uses a cohort only after it contains at least 45 independent station-days. The scanner shows the sample count. Archived forecasts come from the Open-Meteo Previous Runs API. Outcomes come from RCC ACIS station daily summaries.
Venue buckets resolve on whole-degree climate reports. A 91-92 bucket means the rounded value lands in 91 or 92, so the continuous temperature interval is [90.5, 92.5).
Edge
Edge is fair value minus the executable price at the touch, net of fees. YES uses the ask. NO uses 100 minus the bid. That keeps wide spreads from showing as fake opportunities.
A 47 cent edge does not mean you make 47 cents per dollar if you are right. If you buy YES at 35 cents and it resolves YES, the gross payout is 65 cents per contract. The 47-cent edge means the model fair value is about 47 cents better than the current executable price.
Model performance
The useful version is model performance by lead time: for example, how often GFS picked the right bucket 24 hours before close, or whether ICON was better six hours before settlement. The Performance tab scores archived forecast snapshots against official-station observations and reports exact bucket hit rate, within-one-bucket hit rate and average temperature error. It does not fill missing history with simulated model rankings.
Not financial advice
Weather-market edges can disappear quickly. Check the venue book, station observations, source age and market rules before risking money.