How accurate were we?
Nothing to score yet, and that is the expected state until October.
Every Monday from the start of September, the peak window this site was predicting for each location is written into a log that cannot be edited afterwards. A call cannot be graded until a location actually peaks, so this page fills in through October and is complete when the season ends in early November.
It is linked from the homepage now, before there is anything flattering on it, because publishing the promise and the result from the same URL is the whole point.
0
Frozen calls on record
–
Inside the window
–
Median miss, days
The table that will matter
Calling a peak three days out is easy. Calling it in late August is the thing worth judging us on. The layout is reserved; the rows fill in as locations peak.
| Lead time | Calls | Inside the window | Mean miss |
|---|---|---|---|
| 0 to 7 days ahead | – | Scores in October | – |
| 8 to 14 days ahead | – | Scores in November | – |
| 15 to 28 days ahead | – | Scores in November | – |
| More than 28 days ahead | – | Scores in November | – |
Scoring rule · the real peak is the middle of the run of days Ontario Parks reported a location at peak, not the first day, because using the first day would reward a forecast that ran early and clipped the leading edge. A call counts as a hit only if the real peak fell inside the window we published; landing inside the confidence band is reported separately and is a weaker result. Calls frozen after a location had already peaked are excluded, and locations with no Ontario Parks coverage are left out rather than graded against our own ground reports. Full method. Writing about this before then? The press page has the current forecast, and the frozen log is available on request.