How good is the number?
This is our first winter. We don't have a season of results, and we won't invent a percentage.
How accurate is the snow day calculator?
We don't know yet, and we won't put a figure on it until a winter of school mornings can back it up. Anyone quoting an accuracy figure without showing the predictions behind it is guessing.
Until then, read the percentage as what it says. A 30% chance of closure means schools open on most mornings that look like this one and close on some of them.
What we keep
- Every prediction
We save each prediction as it's published: the number, the reasons, and the time. We don't edit it or recompute it afterward.
- The 8 PM and 4 AM calls
For 122 districts and boards, and every place someone has an alert for, we also save a prediction at 8 PM and 4 AM local time each school night, with the full forecast it was made from. Same hours every night, so any later comparison is like for like.
- The version
Every prediction records the version of the method that made it. When the method changes, last week's predictions keep the version they were made with.
How we'll check it
- Your answers
On mornings when school was in doubt, your brief asks whether you got a snow day. Answers tell us where the number ran high or low. They're self-reported, so they tune the model but never count as proof on their own.
- Storm recaps
After big storms, we'll compare our 8 PM calls in the storm's path with the closings that were actually announced, and publish what we got right and what we missed.
How the number is built
The signals, the hours that decide it, and where the method runs out are on how we calculate this.