The past 30 days: what the model would have forecast the evening before, compared with the Torbole observations.
The model was correct on 23 of 25 days with a definite forecast (92%; MAYBE days are not counted as decisions).
The model was correct on 20 of 22 days with a definite forecast (91%; MAYBE days are not counted as decisions).
The model row is a genuine hindcast: it was calculated exclusively from the archived model run issued the previous day, using the same coefficients as the live forecast. The observed row comes directly from the Torbole station’s 10-minute data (unvalidated telemetry).