What the House Knows · Part 1 of 3
Two Identical Bedrooms. One Won't Stay Warm.
In short
- Two same-build bedrooms on the same automation: the north-east one needed 48 per cent more heating over fourteen winter nights.
- The heaters are twins; the difference is heat loss, with the NE room taking 1 h 07 m to lose 1.3 °C against 1 h 35 m next door.
- Wind and afternoon sun are ruled out, leaving a steady loss path with the ceiling insulation as prime suspect.
- 48%more heating in the NE room
- 40%faster heat loss, like for like
- 22%faster loss even on no-sun nights
We have two bedrooms that are, on paper, the same room. Same dimensions, same construction, both on the north edge of the house, one in the north-east corner and one in the north-west. Same new double glazing, same honeycomb blinds. Each has its own heat pump head unit, driven by the same automation: hold the room in a band about a degree wide overnight, with a small temperature probe on the wall as the referee.
And yet. All through June I kept half-noticing that the north-east room’s heater seemed to be running every time I walked past, while the north-west room’s barely murmured. The kind of observation that normally dies in the hallway — huh, that’s odd — and is forgotten by the time the kettle’s on.
Except this house logs everything. So one evening, cuppa in hand, I asked the obvious question: is the north-east room actually using more heat, or am I imagining it?
The data nerd holdings
Every room has a small wireless temperature probe, twenty-odd dollars each, reporting every few minutes. On the roof there’s a weather station: outdoor temperature, wind speed and direction. All of it lands in Home Assistant, which writes every reading, and every time a heater turns on or off, into a database on a mini PC in the study. About five million rows a day. I keep all of it.
Ninety-five per cent of this data is wallpaper. Nobody will ever look at it. But you don’t know in advance which five per cent will matter, and when a question like “is that bedroom really colder?” turns up, the answer is already on disk, for both rooms, with the weather alongside. You cannot reconstruct that after the fact. You can only have been recording it.
Round one: who’s burning more heat?
Fourteen winter nights, both rooms held in the same band by the same automation:
| NE bedroom | NW bedroom | |
|---|---|---|
| Times the heater cycled on | 64 | 39 |
| Total minutes of heating | 821 | 554 |
So no, not imagining it. The north-east room needed 48 per cent more heating to hold the same band on the same nights. Roughly an extra twenty minutes of compressor time every night, for the same result.
Suspect one: the machinery
First and most boring explanation: a tired head unit, weaker, so it runs longer to do the same job. The data says no. Each head climbs through the band at almost the same speed: 8.1 °C an hour in the NE room, 8.5 in the NW. The heaters are twins. Whatever is different about these rooms, it isn’t the machinery pushing heat in.
Which leaves the other side of the ledger: how fast the heat leaves.
The cleanest number in the case
The comfort band makes each room repeat the same experiment all night: heat to the top of the band, switch off, drift back down to the bottom. The drift is a pure measurement of how fast the room loses heat, and each one covers the same 1.3 °C fall. Heat loss scales with the gap between inside and outside, so I only compared drifts where that gap matched, a ten-degree night against a ten-degree night. Otherwise you’re comparing weather, not rooms.
| Time to lose the same 1.3 °C, heater off | NE bedroom | NW bedroom |
|---|---|---|
| Median | 1 h 07 m | 1 h 35 m |
The north-east room loses heat around 40 per cent faster than its supposedly identical twin. The extra heating isn’t preference or thermostat placement or a weak heater. The room can’t hold what it’s given.
The question is why.
Suspect two: draughts
My first instinct was a draught, a failed seal letting the wind in. Wind-driven leakage gets worse when the wind blows, so I lined up every cooling drift against the wind record.
Nothing. Correlation effectively zero, for both rooms. Calm nights and windy nights, the NE room bleeds heat at the same rate. It was a calm fortnight, so a leak that only shows in a gale isn’t ruled out. But whatever is taking the heat is taking it steadily, hour after hour.
Suspect three: the sun
Then the good objection: the NW room gets the afternoon sun. The “better” room may simply be banking free solar warmth every clear afternoon and coasting on it overnight. Maybe the NE room isn’t leaky. Maybe the NW room is cheating.
Three tests, using the solar panels’ output as a sunshine meter:
Does the NW room’s head start survive cloudy days? It does start ahead, about 0.9 °C warmer at five in the afternoon. But on the two most overcast afternoons, when there was no sun to bank, the gap was +1.1 and +0.9 °C, right on the median. Stored sunshine can’t explain a lead that’s undiminished when there was no sunshine to store.
Does the gap fade overnight? Banked solar warmth drains away within a few hours of sunset. The NE room’s disadvantage grows through the night, about 13 per cent worse in the evening, 22 per cent worse between 1 am and 7 am. Whatever afternoon sun did eight hours earlier, it isn’t doing it at three in the morning.
Compare only the no-sun nights. After overcast afternoons, with the sun out of the equation for both rooms, the NE room still loses heat 22 per cent faster. That’s the fabric-only comparison.
One funny thing fell out of the sun check: the only solar effect visible anywhere belongs to the cold room. The NE corner catches the morning sun and on clear days banks a little of it. The sun is subsidising the leaky room, so the real gap is, if anything, slightly bigger than the headline numbers.
The verdict
The loss is real, 30 to 40 per cent faster depending on the measure. It isn’t the heater, the windows, the wind or the sun. What’s left is a steady, always-on loss path that one room has and the other doesn’t.
The prime suspect is the ceiling: insulation that’s missing, slumped, or folded back after an electrician’s visit and never returned. It’s the largest surface in the room, and a bare patch of plasterboard leaks heat around the clock, windlessly, in exactly the pattern the data shows. Second suspect, same signature: air leaking up through downlights or a vent. Either way, the evidence points up.
The data can’t see inside the ceiling. What it has done is eliminate the machinery, the windows, the wind and the sun, and tell me where to point the torch. The next step is low-tech: get up in the roof space over the north-east bedroom and look, or point a thermal camera at the ceiling from below on a cold morning, where a bare patch glows like a bruise. If I find and fix a gap, the drift time gives me a ready-made before-and-after test. The room should go from losing its band in 67 minutes to something like its twin’s 95.
No machine learning, no specialist kit. Arithmetic, patience, and a database that was quietly writing everything down before I ever needed it.
The house has been talking the whole time. This is what happens when you keep the transcripts.