Cold Comfort: Instrumenting a Victorian Terrace to Test What Heat Pumps Actually Deliver
Every heat-pump argument in the UK eventually reduces to the same anecdote: a Victorian terrace, a cold snap, and a householder who says the radiators never got hot. The engineering underneath that anecdote is tractable, but it is rarely made explicit. So we built a rig to make it explicit.
This is not a show home. It is a guarded test enclosure with a calibrated heat emitter loop, a small monobloc air-source heat pump on the other side of a partition, and enough instrumentation to separate what the machine does from what the building does. The aim was to model a typical pre-1919 mid-terrace with a design heat loss in the region of 3.2kW at an external temperature of -3°C, and then to see which design choices actually moved the coefficient of performance.
Constructing something that behaves like a terrace
The enclosure was built to reproduce the thermal characteristics that matter: a solid nine-inch brick external wall on one face, an uninsulated suspended timber floor, a single-glazed bay, and a deliberate amount of background leakage. We did not attempt to replicate the full geometry. We did attempt to replicate the physics — U-values, thermal mass, and air permeability — because those are the variables that determine how a heat pump is asked to behave.
The emitter loop was fitted with a range of radiators sized according to standard BS EN 442 output tables at ΔT 50, which is how most retrofit radiator schedules are still produced. We then ran the same loop at lower flow temperatures and logged the delivered output.
The baseline is fabric, and the fabric is poor
A solid-walled terrace with single glazing and an uninsulated floor will typically sit around 1.8 to 2.2 W/m²K for the walls, with air permeability in the 8 to 15 m³/h·m² range at 50Pa. That is the number that drives everything downstream. It is also the number most often estimated rather than measured.
Fabric improvements remain the cheapest kilowatt-hour in the house, but the practical options in a Victorian terrace are constrained. External wall insulation is frequently refused on conservation grounds or defeated by bay geometry and stone detailing. Internal wall insulation costs floor area and introduces interstitial condensation risk if it is specified carelessly. Suspended floor insulation is disruptive and rarely done at scale. This is why the retrofit problem is hard: not because the physics is unknown, but because the building stock resists the obvious remedy.
Flow temperature is the whole game
The headline finding will not surprise anyone who works in the sector. Delivered heat output falls as flow temperature falls, and COP rises as the temperature lift across the compressor falls. The two curves cross somewhere inconvenient.
At a 55°C flow and 45°C return, our emitter set delivered comfortably more than the modelled design heat loss, and the heat pump ran stably with modest defrost activity. At 45/40 the radiators delivered noticeably less, and at 35/30 they fell short of design conditions altogether. The COP improved at each step down, but only because the building was being asked to accept less heat per radiator.
Measured at steady state, with a heat meter on the emitter loop and a separate electricity meter on the outdoor unit, the results were as follows:
| Flow/return (°C) | Emitter output (kW) | Measured COP at 2°C ambient | Declared SCOP at 35°C flow |
|---|---|---|---|
| 55/45 | 3.6 | 2.4 | 4.3 |
| 45/40 | 2.8 | 3.0 | 4.3 |
| 35/30 | 2.0 | 3.4 | 4.3 |
The declared seasonal figure is not a lie. It is a laboratory figure for a specific flow temperature, in a specific climate, with no distribution losses and no cycling. Our rig sits in a UK outbuilding, has uninsulated primary pipework over a short but not negligible run, and cycles at part load. Those are not exotica. They are the normal conditions of a domestic installation.
Where the guidance goes soft
The fuzziness starts with heat loss. A room-by-room calculation performed properly is a different document from a whole-house estimate divided by floor area, and yet both are presented to homeowners under the same heading. If the heat loss is wrong, the radiators are wrong, and if the radiators are wrong the installer compensates with a higher flow temperature — which quietly erases the efficiency that justified the installation.
The second soft spot is the word 'low temperature'. A system designed for 45°C flow and a system designed for 55°C are both sometimes described as low temperature, and the difference in both emitter cost and running cost is material. The third is defrost. In cold, humid UK conditions a monobloc unit will spend a measurable fraction of its time in defrost cycles, and the seasonal figure rarely reflects how often that happens on a still January morning in the Midlands.
What a householder should demand before signing
Our rig is a model, not a house, and we would not present measured numbers from it as a prediction for any specific property. What it does support is a list of things worth insisting on.
Ask for the room-by-room heat loss calculation, with the U-values used, and check whether the surveyor actually inspected the wall construction or assumed it. Ask for the design flow temperature in writing, and for the emitter schedule at that temperature rather than at ΔT 50. Ask how the cylinder is sized and what happens when the heat pump is also asked to produce hot water on a cold evening. Ask for a heat meter and a separate electricity meter, or at minimum for the monitoring data from the first twelve months, so that the seasonal performance can be calculated from measurement rather than estimated from a datasheet.
None of these requests is unreasonable, and all of them are cheaper than discovering in February that the system only works when the flow temperature is set to a figure nobody mentioned at the quotation stage.
The technology is not the weak point. The weak point is the chain of assumptions between a heat loss survey and a commissioning certificate, and that chain is only as strong as the measurements it rests on.
