Most weather apps never tell you how often they're right. We check every forecast against what actually happened, and those grades decide what you see.
Several times a day we save what each forecast predicts, hour by hour, for a network of points across each city we grade. Several points per metro, not one, because the weather in Boulder and in Parker can be very different on the same day. Our own blended forecast is saved too.
When the time comes, we compare each prediction with official observations: a high-resolution analysis of measured conditions from the National Weather Service, at the same place and hour. We grade temperature, dew point, wind, cloud cover, rain, and each day's high and low.
A forecast that's great in Chicago winters can be mediocre in Denver summers, and every forecast gets worse further ahead. So grades are kept separately for each place, time of year and how far ahead the forecast was made, and we keep them year over year, so each season learns from the same season before.
When you open EZWX, we fetch the forecasts for your exact location and weight each one by how accurate it has been near you, for this season and this lead time. Forecasts that consistently run warm or cold are corrected for it. Recent performance counts more, but a long track record keeps one bad week from swinging everything. Which forecasts we blend, and the exact recipe, is ours, and it keeps changing as the grades come in.
The blended forecast you see is saved and graded exactly the same way. That's how we know whether blending is actually helping, and it's where the numbers in the app come from.
We publish our graded accuracy by city, updated daily: see accuracy by city. When we miss, it shows there too.