Apologies in advance for the long post!

My house has run on a hybrid system since it was built in 2013: a Viessmann Vitocal 200-S air source heat pump working alongside a Viessmann Vitodens 200 gas boiler, installed together on Viessmann controls. The heat pump is type AWB-AC 201.B13, a 13 kW split, with the boiler covering the peak and doing all of the hot water as a design decision.

The heat pump has now reached end of life, with Viessmann confirming that spare parts are no longer available. Even though I had the fan motors and control PCBs replaced three years ago, I am worried that something else will break. Rather than waiting for this to happen, I would like to replace the unit and take the opportunity to remove the boiler at the same time to come off gas completely. I need to find a unit that will carry the entire load on its own and also do the hot water.

That makes the design load figure the whole question and it is where I have run into trouble. I have two answers that differ by a factor of 1.7, and I would like someone to tell me whether I have got this wrong before I commit to equipment.

The house

It is a developer-built detached house completed in 2013 in Surrey, on four storeys including a basement and rooms in the roof, with around 550 m2 of floor area excluding the garage. There is underfloor heating on three floors, radiators on the second and MVHR throughout.

The as-built U-values are 0.28 for the external walls, 0.16 for the pitched roof, 0.21 for the flat roof, 0.22 for the ground floor and 1.8 for the windows. Air permeability is 8.2, according to the EPC.

The data

The hybrid arrangement has given me two distinct measurement regimes, which is the reason this post might be useful to anyone else working a design load out from consumption figures.

The heat pump was switched off entirely from 2014 until February 2021 and the boiler carried the whole load, which gives a clean gas-only record; my meter readings cover 2015 to 2018 within that period. Since 2024 I have had daily gas consumption metered separately for hot water, along with heat pump output logged into InfluxDB, which gives a period where both heat sources are running and can be separated from each other.

Three methods

Firstly, I regressed twenty gas meter intervals from 2015 to 2018 against heating degree days computed hourly from ERA5 data at my coordinates. The hot water baseline was taken from each year's measured summer consumption rather than assumed, and the fit was forced through the origin so that only one parameter remained free. That gives 815 W/K at a fitted base of 13 C, with an R2 of 0.96.

Boiler seasonal efficiency is taken as 90 per cent gross, which is reasonable for a Vitodens 200 condensing on underfloor return temperatures. It is worth noting that gas meter readings convert on gross calorific value, so the 98 per cent net figure manufacturers quote is not the number to use here. Varying it between 85 and 92 per cent moves the result by about 1 kW.

Secondly, I took the 2024-25 season and combined heat pump output with boiler heat to give total delivered heat. That gives 814 W/K.

Thirdly, and this is the one I trust most, the heat pump saturates at around 7 kW below about 5 C and its output barely rises as the weather gets colder, so the boiler supplies almost all of the additional load. That makes the slope of boiler output against outside temperature a measurement taken entirely from the gas meter, with no heat pump instrumentation involved at all. The boiler contributes 538 W/K and the heat pump 276 W/K over the same weeks, giving 814 W/K combined. This is only possible because it is a hybrid, and it is the strongest evidence I have.

Where the calculator disagrees

The de Podesta HTC estimator gives 474 W/K and a 10.2 kW heat pump from the same consumption figures.

The entire difference comes from one assumption. The estimator places the balance point 3.5 K below internal temperature, which for me gives a base of 18.5 C and roughly 2,550 degree days a year. Fitting the base to the data instead gives 13 C and about 1,350 degree days. Dividing the same annual consumption by half as many degree days doubles the coefficient, so neither of us has made an arithmetic error, we have simply used different denominators.

The tiebreaker

In the week ending 13 January 2025, at a mean outdoor temperature of 0.8 C, the house consumed 12.75 kW continuously for seven days. Of that, 5.01 kW came from the boiler, which is a gas meter reading with no modelling in it whatsoever. The calculator's 474 W/K predicts 8.4 kW at that temperature, so the house measurably used half as much again as the model says it can.

Where I think I am weakest

A base of 13 C against a 21 C setpoint implies around 6.5 kW of free gains, which is considerably more than a normal house. Some of that is real, since there is a lot of south-west glazing and a secondary hot water loop was dumping about 900 W into the fabric around the clock until I put it on demand. But some of it is almost certainly that not every room sits at the thermostat setting, so 21 C describes the hall rather than the basement or the roof rooms.

Another thing worth knowing is that Viessmann's energy balance is calculated from temperature sensors rather than metered, and Viessmann themselves state it can deviate by up to 20 per cent. The bias is not random: calculated heat output tends to read high while calculated electricity reads low, so the resulting COP flatters and several owners have logged discrepancies of 30 per cent against calibrated meters. That is why the third method matters. If my counter overstates output by 20 per cent, the answer falls only to 760 W/K and 18.3 kW, because the heat pump contributes just a third of the fitted slope.

Where I have landed

The heat transfer coefficient is somewhere between 760 and 815 W/K, which gives a design load of 18 to 19.6 kW at -3 C on the MCS basis that ignores gains, or 14 to 16 kW if you credit the gains the house actually has.

I am specifying delivered output of 15 to 16 kW at A-3 / W37, to be evidenced from the manufacturer's capacity table rather than the A7 / W35 headline rating. Hot water is a comparatively small addition on top: around 10 kWh a day into a 200 litre cylinder at 48 C, which I intend to schedule on Agile Octopus cheaper rates rather than reheat on demand as the boiler currently does.

There is one piece of good news in all this. My weather compensation curve delivers 35.5 C flow at design conditions, which is low for a house this size. In 2018, running on the boiler alone, the curve was set to 54 C to hold the house one degree warmer than it is now, but that was using room controllers to regulate the temperature and I am not doing that any more.

Questions

I know there is quite a lot to take in here! But does anything above look wrong to you?

I don't think there are any monoblocs on the UK market that deliver 15 to 16 kW at A-3, so this looks like two units in parallel sharing the load. I should be clear that I mean parallel rather than a two-stage refrigerant cascade, which is what usually comes up when you search the term. Has anyone here actually run one domestically, and would you do it again? I have read the Grant Aerona 2x10 thread, which is largely about when the second unit should cut in, and that is precisely the part I would want designed rather than guessed at.

I am also open to being told that keeping the hybrid is the sensible answer at this size, but I would rather be rid of the gas and the standing charge that goes with it, and the boiler is thirteen years old anyway.

Separately, is anyone running active cooling into an MVHR duct cooler? I have that now and want to keep it,.

Full disclosure: I did the regressions and the degree day work with Claude, which pulled the hourly ERA5 data, ran the fits across a range of base temperatures and did the sensitivity analysis. The meter readings, the heat pump logs and the U-values are all mine and measured, and the arithmetic is all checkable. I mention it because the conclusion here contradicts a widely used spreadsheet and I would rather people know how it was produced and pick holes in it than take it on trust. I am happy to post the raw data, the workings or the code if anyone would like to.