The weather service's forecast, scored.
How close the National Weather Service's own published forecast came to the high or low each Kalshi temperature ladder settled on, city by city. A measurement of the government's forecast, kept beside the market for reference: no prices, and nothing here is advice on any contract.
reading…
The published forecast's record
Daily high · published forecast, 24 hours before the close
—
reading…
Daily low · published forecast, 24 hours before the close
—
reading…
The high is published for the day 24, 12 and 6 hours before the close (the 6-hour figure only when it was fetched before 6 pm local time); the low only at 24 hours: its night period, already under way by then, ends on the date's morning and has left the forecast by 12 hours before the close, so it covers only the day's first hours.
City by city, 24 hours before the close
What was known by then: the hourly forecast beside the day's readings
Later in the day the record also keeps the weather service's hourly forecast for the hours still to come, set beside the day's own readings so far: the higher of the two for a high, the lower for a low. By 12, 6 and 1 hours before the close the readings carry more and more of it, so these rows say what was known by then, not how good the forecast was. At 6 hours most highs are already in the readings. They are kept apart from the published forecast's figures above, and not compared with them.
How it is measured
The settled reading. Kalshi settles each ladder on The Weather Company's reported high or low for the station its rules name. The record keeps which bracket settled Yes, which is two degrees wide (or an open tail), and the station's own day from the government's observations, rounded as the climate report rounds. The bracket decides and the station gives the degree inside it: where the station's reading lies outside the bracket that settled Yes, it is moved to the nearest degree inside that bracket, and those days are counted. A ladder not yet settled, or a day with no station reading, is left out.
The forecast. The weather service's 12-hour forecast as the worker fetched it, at the last fetch at or before 24, 12, 6 and 1 hours before the close: the daytime period's high for the date, and the night period's low that ends on the date's morning. The close is the end of the station's day, midnight local standard time.
The figures. The error is the forecast minus the settled reading, in °F. "Ran warm" or "ran cool" is its mean: positive, the forecast was warmer than the settled reading on average. The mean miss is the mean of its size, whichever way. Within 1°F and within 2°F are the shares of days the forecast was that close. The days are city-days: one city's high or low on one market day.
Few days yet. Under 30 days a city's figures are a record of those days, not a property of the city: one 3-degree miss over 12 days moves its mean by a quarter of a degree. The cities are listed in order of their names and never ranked.
Gaps. On a few nights the database lost the worker's saves, so no 24-hour forecast was kept for some days in some cities (for 2026-10-17, 2026-10-18 and 2026-10-20 in the cities outside the Pacific zone, and 2026-10-21 in the Eastern ones). Those days are missing here, not counted as misses, so the cities' day counts differ.
A measurement of the weather service's forecast, not a forecast of the site's own, and not advice on any contract.
Where these numbers come from
| Part | Read from | Refreshed |
|---|---|---|
| Every figure on the page | nws_: per city and over every city, daily high or low, kind of forecast and hours before the close: the days, the mean error, the mean miss, the shares within 1°F and 2°F, the days moved into the settled bracket, and the first and last day | on every read (a plain view over the record) |
The National Weather Service (api.weather.gov): the 12-hour forecast and the hourly forecast for each station, fetched by the worker every pass, and the station's observations, which give the day's high and low. Kalshi: which bracket of each ladder settled Yes. The Weather Company: the source Kalshi settles on; this page does not read its figure, only which bracket Kalshi settled.
The view is defined in nws_forecast_scores.sql, over the worker's tables of forecasts, station days, brackets and results: counts, means, shares and the first and last day only. No prices are read or published, and nothing here is offered as a download.