The Switching Question · Part 1 of 7
The Curtailment Trap: Why My Own Bills Nearly Fooled Me
In short
- My summer bills were accurate, but they recorded how the house behaved under a wholesale plan, including deliberately curtailing exports when prices went negative.
- By the model's reckoning, about 40 per cent of the panels' potential summer generation never happened.
- Any plan that rewards exports looks worse than it is when it is judged on that curtailed history, so the comparison needs potential generation, not observed generation.
- 40%of potential summer generation never happened (modelled)
- 19.4 kWsolar array
- 48 kWhbattery
I almost switched electricity plans on the strength of one summer of my own bills.
The bills were accurate. Every reading traced to the meter, every charge to the price that applied. By any normal standard they were the best evidence in the house. They were also the wrong evidence, wrong in a way my later modelling priced at hundreds of dollars a year, while the comparison built on them looked rigorous the whole time.
The culprit is curtailment: choosing to waste your own sunshine because on certain afternoons exporting it would cost you money. My house did this deliberately. It was the right call under the plan I was on. It also quietly rewrote my generation history.
Exhibit A: my own bills
The standard advice for comparing plans is to start with your own usage history. Download the data, run it through each plan, see which would have cost least. It feels unimpeachable: your actual house, your actual habits, your actual summer.
I did exactly that. I took a summer of half-hourly readings, every kilowatt-hour bought and sold, and worked out what the same summer would have cost on each plan I was considering. (A kilowatt-hour is roughly one bar of an electric heater running for an hour.) The arithmetic was careful and the data was genuine, and the ranking looked authoritative enough to act on.
The hidden assumption is that your history is a neutral record of what your house does. For a house without solar that’s close enough: the kettle doesn’t care what electricity costs. With solar and a battery it isn’t true at all. Your history records how you behaved under the plan you were on. Change the plan, and the behaviour changes, and the data with it.
Choosing to waste sunshine
Last summer I was with Amber, on a plan that passed the wholesale electricity price through to me. That price moved every half hour, and so did my feed-in price, the money I got for each kilowatt-hour I exported.
On plenty of summer afternoons it went below zero. Not low: negative. For those hours, exporting solar wouldn’t earn me money; it would cost me.
Why wholesale prices go negative is a question this series leaves alone. For this story, negative prices are weather: a condition that exists on certain afternoons, and that the house responds to.
The response is simple. My home automation watches the price, and when exporting would cost money, it stops. The battery is usually full by then, so the system throttles the panels back to what the house can use in the moment. On a bright February afternoon, that is a small fraction of what they can produce. The rest of the sunshine lands on the roof and isn’t harvested.
Nothing is broken and no cloud has rolled in. The panels produce less because producing more would lose money under this plan. It is a response to a price, not a property of the solar system.
The forty per cent hole
I didn’t think much about the gap for a long time, because nothing draws attention to it. The bills don’t show it, the monitoring app doesn’t, and the meter records what the panels produced, not what they could have. Power you chose not to generate leaves no paperwork.
The gap only surfaced when I built a model of potential generation: an estimate, from local weather records, of what the panels could have produced each day if nothing had held them back. The next post covers how it was built. What matters here is what happened when I laid it over last summer’s meter readings.
Roughly 40 per cent of my panels’ potential summer generation never happened.
I made a pot of Yorkshire tea and sat with that for a while. It is not a rounding error, and it is not a few hazy afternoons. It is two-fifths of the summer’s sunshine, deliberately declined by an automation I had set up myself, doing exactly what I’d asked.
The gap is widest on the sunniest afternoons, which are exactly the afternoons the export price goes negative. The days my system could generate the most were the days it was told to generate the least.
Same roof, two histories
Take the identical house, same panels, same battery, same weather, same people forgetting to turn the bathroom light off, and put it on a flat-rate plan with a fixed feed-in price.
There, every kilowatt-hour earns the same whenever it’s sold, so there’s never a reason to stop exporting. The automation that curtailed on my wholesale plan would never fire. The meter would record something very close to full potential.
Same roof. Two plans. Two completely different generation histories. My observed generation isn’t a measurement of my solar system alone. It’s the system entangled with my behaviour under one set of prices. The number that looks like physics is partly choices.
What this does to a plan comparison
When I priced the candidate plans against my summer history, every plan that rewards exports started from a hidden handicap. My history says this house has roughly 40 per cent less summer sunshine to sell than it really has, so a plan paying decently for exports was marked down for exports I had chosen not to make, a choice I’d un-make the moment I joined it.
That’s the trap, and it’s circular: behaviour optimised for plan A, used as evidence to judge plan B. The better my automation got at playing my wholesale plan, the worse a witness my data became for any other.
I want to be precise about what the bills got wrong, because the answer is nothing. The bills aren’t wrong. They are exact answers to “what did this house cost under this plan, behaving this way?” But I was asking what it would cost under a different plan, and under a different plan the history itself would have been different. The would-have-been bills I’d computed assumed I’d keep behaving as though I were still on the old plan, which is the one thing nobody would do.
Manufacturing the missing evidence
The comparison needs better evidence, not better arithmetic, and that evidence isn’t on any bill. It needs potential generation, not observed: the sunshine that actually arrived, converted by the panels I own, with no automation in the way. That has to be built from weather records, checked against the meter, and tested across more than one summer so a freak season can’t skew it. That’s where the series goes next: something like manufactured hindsight, a fair basis for comparing plans that my own bills could never be.
If you have solar and you’re comparing plans on your own history, the question to ask isn’t whether your data is accurate. It’s what your data is a history of. If you’ve ever held back exports, shifted loads to chase a cheap window, or let a battery soak up power that would otherwise have been sold, your history is partly a portrait of your old plan. The better you played that plan, the worse a witness it is.
The bills were honest. The sample was rigged. And I was the one who’d rigged it.
The spreadsheet doesn’t care what you hoped the answer was.