Forecast accuracy
Every forecast we publish is checked against the observed temperature at the station that settles the market.
Published since October 20, 2026
Average error: mean absolute gap between our central forecast and the observed temperature.
Error vs raw model average: how much lower our error is than simply averaging the forecast models (shown for tomorrow and the day after).
80% range coverage: share of days when the observed temperature fell inside our 80% range. It should stay close to 80%: much more means our ranges are too wide, much less means they are too narrow.
The engine was launched on 20 October 2026: figures are based on few days so far and will settle as data accumulates.