The record

What this page still needs

1. Time for our own thermometer to catch up. weather_history now records a real daily high and low, widened on every station reading, but it only started doing so on July 30, 2026. 14 of the 14 days above are scored against that reading; the rest are still scored against the Open-Meteo daily maximum over the station's coordinates.
2. A longer archive. Ours starts in May 2025, which is enough to score a fortnight or a month but not enough to say whether we forecast this July better than last July.
3. An observed condition to compare against. No feed we read publishes what the sky actually did, so a forecast of thunderstorms is scored on whether rain fell and nothing else. Rain called against rain fallen is the only condition check on this page.
4. Retention on weather_forecast_history. It takes a row per forecast day per station every ten minutes. That archive is what makes this page possible and it is also the fastest growing table we have.
5. Station normals. The comparisons in the month panel come from a thirty year reanalysis over the station’s coordinates, not from the published climate normals for the station itself, which is why the hot-day comparison is withheld wherever the two count different thresholds.