Model weights stay frozen within each outer evaluation block. Historical hindcasts through 2022, validation comparison in 2023–2024, and previously inspected 2025–2026 results remain separate. Recent errors do not select a preferred model or history. Missing truth stays missing; complete-month targets require every daily observation. Seasonal volume uses the actual number of days in each month.

Only complete April–July sets appear. Missing daily truth prevents an observed complete-season volume.