How wrong is a solar estimate for a real roof?
+8.1 %
MEASURED against MODELLED
73 days, 5 July to 16 September 2026, one house
Every solar quote comes with a number: how many kilowatt-hours the panels will make in a year. It comes from a model — the sky over your postcode, the angle of your roof, a standard allowance for losses. Nobody tells you how far that number can be from what the panels actually do.
This house has a 12.96 kWp roof on two faces (east-south-east and west-north-west) and an inverter that records what it made, every five minutes. So the same model the tool uses was run over the same days, with the real weather over the house, and set beside the inverter's own record.
The model said 3,398 kWh over 73 days. The roof made 3,143 kWh. The model was 8.1 % high.
Day by day the model tracks the roof closely — a bright day is bright in both, a dull day dull in both. The gap is mostly the model reading a little high on clear days, which is what you would expect from a standard loss allowance that was never tuned to this roof, this wiring or this summer's dirt.
The comparison runs every morning at 07:30 on the house's own computer: the inverter's record is frozen for the day, the model is run over the same days with the weather archive, and the difference becomes the band the tool quotes. If the two disagree by more than 10 %, the morning is marked red and nothing is published until it is understood.
On record: the validation results, entry "E3 re-run over the whole record" (14 Sept 2026), and the morning re-check that writes the band the tool quotes. Published 17 Sept 2026.