Every station on one day, each forecast beside the temperature the station recorded. The colour is the error, so a system that ran warm reads as a red stripe down its column and a station nobody caught reads as a red row. The margins are mean absolute error, the right column per station and the bottom row per system. Pick either end of the day and any of the last seven scored days.
The argument sets out why a market price can be a forecast at all, and why it can beat the products it is built on. This page is the measured record behind it.
Every scored day at a single station, with each tool’s error beside its forecast, is on that station’s own page. Reach one from a dot on the map.
NWS is the day/night forecast product; NBM is the National Blend of Models’ own daily max and min; GFS MOS is a single-model statistical product with no forecaster editing; LAMP is observation-updated hourly guidance with a 25-hour horizon. NBM is the guidance forecasters start from when editing the NWS grids, so those two are related, not independent. The ForecastEx line in the figure is the strike where the Yes price crosses 50¢, read from the last quote before local midnight. It is a price, so it is drawn in the figure and left out of the skill statistics. The record begins the day this site’s archive started; it grows by one day every day.