Checking the model against a real Irvine balcony

Why check it at all

A model is a set of assumptions with arithmetic on top. Ours is public and every figure on the site traces to it, which is the whole basis on which we ask anyone to believe a number here.

That only means something if the assumptions get tested against the world occasionally. Most balcony solar content is manufacturer specifications restated, or model output presented as fact, or German figures applied to California without adjustment — and none of it is checkable, because none of it shows its working.

We'd rather not add to that. So: a prediction written down first, a logger, and the gap published either way. This isn't a test lab and it isn't a study. It's one reading against one forecast, which is enough to catch a model that's badly wrong.

The setup

What our model predicts

Before any data, here's the shape of what our methodology page says a balcony like this should do:

1,586 kWh/kW/yr × kW × direction factor × tilt factor × shade factor = annual kWh

The specific numbers go up alongside the kit specification, and they're written down before the first month's readings so we can't quietly revise them later.

What we're measuring

Daily production, which is the headline.

Production against forecast weather, so we can separate "the model is wrong" from "it was cloudy in June."

Self-consumption, which is the harder and more interesting one. Producing 4 kilowatt-hours means nothing if the flat is empty and 2 of them go to the grid for free. We're logging household draw alongside production to work out what share is actually used.

The tilt experiment. At some point we'll run the same panel vertical on the railing for a month and tilted for a month, in comparable conditions, to test the 0.62 factor we've been quoting. If that number is wrong, a lot of our advice about tilt brackets is wrong with it.

Seasonal spread. December production in California is roughly half of June's. Anyone quoting an annual figure from a summer month is overstating by a wide margin, and we want the real curve.

What we'll publish

A monthly update with the numbers, and a proper comparison at twelve months.

Specifically, we'll publish the gap between prediction and reality, whichever direction it runs. If the model overstates by 20 percent, that's the most useful thing we could tell you, and we'll change the model and say what changed.

What this can't tell you

One balcony. One orientation. One year of weather. It'll tell you whether our model is roughly calibrated for a west-facing Irvine balcony, and it won't tell you what happens on a shaded east-facing one in Fresno.

That's the honest limit of a single logger, and we'd rather say it than imply this is a study. It's here to check whether the model is roughly calibrated — not to stand in for the sourcing and arithmetic the rest of the site runs on.

If you have one

If you've got a plug-in system running anywhere in California, we'd like your numbers — kit, orientation, tilt, and monthly kilowatt-hours. A dozen real installations would be worth more than our one, and we'd publish the aggregate with the model comparison.