What the House Knows · Part 2 of 3
The Night the Battery Ran Dry
In short
- A battery that left the free window genuinely full hit its 8 per cent floor at about 9:45 am after the first sub-zero night, drawing about 1.4 kWh from the grid.
- The day's 7 °C average predicted about 19 kWh of heating; the house took 29.8, because heating follows the overnight minimum, not the mean.
- The 8 per cent floor is a chosen blackout reserve, so the 29 cents was the premium on that insurance, not a failure.
- −0.1 °Covernight minimum
- 1.4 kWhgrid draw after hitting the floor
- 29.8 kWhheating taken (19 predicted from the mean)
On the coldest night of the winter so far, my house battery ran out. Not dramatically — no flickering lights, nobody noticed a thing. At about a quarter to ten the next morning, with the battery scraped down to its hard floor, the house quietly reached back out to the grid for an hour or so, until the sun and the cheap window arrived to bail it out.
Total damage: twenty-nine cents.
Almost everything about that night was the opposite of what I assumed while watching the gauge sink. I assumed the system had failed. It hadn’t; it had done what the arithmetic said it must. I assumed the expensive evening peak was the casualty. It wasn’t even close. And I assumed the fix would involve buying something. The actual fix is free, and the main reason to bother isn’t money at all.
The shape of a winter night
The battery’s job in July is nothing like its job in January.
My electricity plan has a free window: eleven till two every day, when anything I draw from the grid costs nothing. In winter the solar panels produce a fraction of their summer output, so the battery leans on that window hard. At eleven it charges at full tilt and by early afternoon it’s full. From two it’s on its own: it carries the house through the expensive three-till-nine evening peak, then through the night, and has to still be going when the window opens again at eleven. Twenty-one hours on one charge.
Most winter nights, that’s comfortable. The overnight drain runs at about two percentage points of charge an hour — steady, metronomic, mostly the heating. The battery lands the next morning with charge to spare.
Mostly.
The night in question
Then came the night of 9 July, the first this winter the thermometer on the roof went below zero, bottoming at −0.1 °C before dawn.
| Time | Battery | What’s happening |
|---|---|---|
| 1:30 pm (9th) | 100% | Free-window charge done. |
| 9 pm | 78% | Evening peak ends. Barely dented. |
| 4:30 am (10th) | 39% | Heating has been grinding all night. |
| ~9:45 am | 8% — floor | The battery’s minimum. It stops discharging. Grid takes over. |
| 11:00 am | — | Free window opens; battery recharges at full rate. |
| ~1:30 pm | ~100% | Back to full. As if nothing happened. |
That 8 per cent is the discharge floor I set myself. When the battery got there, the house fell back to the grid for about an hour and a quarter, drawing about 1.4 kilowatt-hours before the solar ramped up and the free window finished the rescue.
Over that one charge the battery delivered about 44 kilowatt-hours, and roughly three-quarters of the overnight drain was heating. It started the night genuinely full. There’s no “if only it had finished charging” escape hatch in this story.
Was something broken?
My first reaction was that something had misbehaved, a heater left running, an automation gone rogue. This house has form on that front.
So I pulled the thread. The boring, satisfying answer is that nothing misbehaved. It’s a straight line.
Eighty-five days of winter data say heating demand follows outdoor temperature with a simple rule: below about 15 °C, every degree colder costs about 2.4 kilowatt-hours a day in extra heating. The heaters don’t meter themselves, so “heating” here is what the house drew above a mild day’s baseline. A day averaging 10 °C needs roughly 13 kWh of heat; a day averaging 5 °C needs about 25. The days queue up along that line obligingly.
A full battery holds a fixed number of kilowatt-hours. Heating demand rises as the mercury falls. Somewhere those two lines cross, and beyond the crossing the battery cannot make it to eleven. Not won’t. Can’t. The night of 9 July was the first on the wrong side of the line.
The average that lies
By the daily summary, 10 July doesn’t look extreme. The average was about 7 °C, cold but nothing this winter hadn’t served up before. Fed into the straight-line model, that predicts about 19 kWh of heating. The house took 29.8.
The model wasn’t wrong; the average was. Heating is consumed hour by hour, and the brutal hours cluster in the small of the night, where the minimum lives. A day that averages seven by being mild at lunchtime and sub-zero at 5 am is a different animal from a day that sits at seven all day. Same mean, very different night.
The lesson: for overnight loads, the daily minimum is the number that matters, and the daily mean will lie to your face.
So where’s the line?
The nights either side of the cold snap, with minimums between +1 and +3 °C, all landed with 18–27 per cent still in the tank. The −0.1 °C night landed on the floor. The trend crosses the floor a whisker below zero.
So for this house: if the overnight minimum is going below about 0 °C, the battery will not reach the free window. Above about +2 °C, it’s never been close to trouble.
That threshold hangs off exactly one floored cold night. I wouldn’t defend the decimal.
The twenty-nine-cent catastrophe
The money says: shrug. The grid draw was 1.4 kWh at off-peak rates, twenty-nine cents. This corner of Melbourne manages only a handful of sub-zero minimums in a typical winter, so letting this happen costs three to six dollars a year. There is no gadget on earth I can’t resist buying for less than six dollars a year, but the spreadsheet is firm on this one.
The peak was never in danger. My instinct said running dry meant the expensive window was at risk, the one the whole system is built to defend. The battery ended the peak at 78 per cent. The place it actually runs dry is four to ten in the morning, the unglamorous off-peak shoulder where the heating has been chewing through charge for hours and the sun hasn’t turned up for work yet.
Holding back more reserve is the wrong fix. The obvious knob — raise the 8 per cent floor to keep more in hand — does precisely nothing here. The battery already starts the night full. A higher floor doesn’t reduce the shortfall; it just moves the same grid import earlier in the morning. This one took me an embarrassing tea’s worth of staring to accept.
The right fix is free. The house can bank energy somewhere other than the battery: in the rooms. The free window costs nothing, so on days when the forecast overnight minimum dips below about +2 °C, the plan is to run the heating a little harder inside it and walk into the evening a degree or so warmer. Every kilowatt-hour of warmth banked at midday is one the battery doesn’t spend at 3 am, and about 3 kWh of margin would have kept the battery clear of its floor even on the worst night so far. That’s small enough to pre-bank as heat. The rooms hold pre-heated warmth right through the evening peak without any top-up. The thermal mass is a second battery that was there the whole time.
The real reason to bother, given the six-dollar stakes, is the battery’s health, not the bill. Scraping a lithium pack to its floor on every cold snap is the habit the chemistry dislikes. Pre-heating costs nothing and spares the pack a deep discharge. That trade I’ll take.
The biggest one: the battery never actually ran dry. That 8 per cent floor is about 4 kilowatt-hours I keep locked away all year as blackout insurance, enough to run the fridge, the lights and the internet for a decent stretch if the grid fails. The morning’s grid draw was 1.4 kilowatt-hours. The house didn’t exhaust the battery; it ran into my own fence, with nearly three times the morning’s import still sitting behind the wire. Drop the floor to zero and this post doesn’t exist.
So the twenty-nine cents wasn’t the cost of a failure. It was the premium on an insurance policy, paid only on the handful of nights each winter cold enough to press the house against its own reserve, for cover that stands guard every night of the year. And that threshold isn’t a law of physics either: it’s where the weather meets a policy I chose. A lower floor slides the line colder. The thermometer proposes, the floor disposes.
Framed that way, I’d pay the twenty-nine cents twice.
The other dry night
The battery had floored once before, a month earlier, on a night that wasn’t even cold. Different disease entirely: that day the free-window charge had only reached 35 per cent before the window shut, and no battery survives the night on a third of a tank. One’s a demand problem, the other’s a supply problem. The gauge tells you that it died; only the history tells you why.
No machine learning. A ruler, mostly.
The house has been talking the whole time. This is what happens when you keep the transcripts.