Learned from the last 30 days of (POA-modelled kWh) ÷ (actual DC kWh), bucketed by cloud-cover %. Likely shedding hours (SOC≥99% & no export) are excluded so the curve reflects what the panels could deliver, not what the inverter chose to absorb. Last run 19 May 16:16.
Solid amber = expected solar irradiance (W/m²) at the panel orientation; violet = ambient temperature (°C). Cloudy hours visibly knock down the irradiance.
Hourly kW after Hay-Davies POA, site cloud calibration, NOCT cell-temp derating, and your SystemEfficiency × kWp.
Today is excluded. Walking back from yesterday, the page picks the most recent 7 days where
the inverter was online for at least 22 of 24 hours (frames present for
TotalDCInputPower); incomplete days (storm outages etc.) are skipped and the next earlier day
is fetched instead. For each picked day, Open-Meteo's archive supplies observed irradiance, the same kWh
model is applied (MJ/m² × 0.278 × 10.54 kWp × 0.72),
and compared to daily_summary.production_kwh.
Skipped: 12 May (system offline ≥ 2 h).