How each widget works

KarbonKit’s models match the outside dimensions of real products of each size, without their branding. They are fixed at full size, so pinching doesn’t resize them.

The phone finds the floor and places the model on it (WebXR on Android, AR Quick Look on iPhone). Placement is more accurate in good light, on a floor with some texture.

The recording is of a real heat pump at full output, which is louder than it usually runs. Visitors set the volume to match the unit they are considering.

The meter averages the microphone signal and adds 94 dB to estimate a sound level. Phone and laptop microphones aren’t calibrated, don’t apply the A-weighting that noise limits use, and adjust their own sensitivity, so readings can be out by 10 dB or more. The meter is most useful for comparing one place with another.

A metre away in the open, a heat pump is about 8 dB quieter than the sound power on its data sheet. It gets about 6 dB quieter each time the distance doubles.

The house is an illustration, not a design: a typical 1930s semi with a heat pump, hot water cylinder, radiators, solar panels, a battery, a car charger and an induction hob.

Costs, savings, performance and noise are given as typical ranges. Each topic lists its sources, with the date they were checked, under “Sources”. A text-only version lists everything and works on any device.

Most sources are gov.uk, Ofgem, the Energy Saving Trust, MCS, DESNZ statistics and the Electrification of Heat trial.

The directory comes from Companies House, public business listings and installers’ own websites. Results are sorted by straight-line distance from the postcode. Installers can’t pay to be listed higher.

The technologies each installer offers are taken from their website where possible. Each card shows where this came from: their website, brands they mention, or their company name. The last two are marked as not confirmed.

Each week, the records that have gone longest without a check are checked again: company status at Companies House, whether the website works, and whether it still lists the same services. Dissolved companies are removed.

A website owner can list their own suggested installers instead of the directory, or above it. These are labelled as suggested by that website.

Each scheme’s rules (country, tenure, income, benefits, property and heating type) are stored as conditions and checked against the answers. If a required condition fails, the result is “Not a match”. If every condition passes, it is “Looks like a match”. Otherwise it is “Worth checking”, with the points to confirm listed – for example when a question was skipped, or a rule needs more detail than four questions give.

The postcode gives the country and council. In England it is also checked against the government’s list of postcodes eligible for the Warm Homes: Local Grant.

Schemes past their closing date are hidden. A scheme that is overdue for review, or has an announced change we haven’t added yet, shows a warning to confirm the details on the official page.

Heat demand is the heat the home needs each year. Without an address, the widget starts from typical fuel use and multiplies it by the boiler’s efficiency. With an address, it uses the home’s EPC instead: floor area times the median gas use per square metre of houses with the same rating (from meter readings), converted to heat in the same way.

Keeping the boiler: heat demand divided by boiler efficiency gives the fuel burnt, which is multiplied by the fuel’s carbon factor. This is the same every year.

Switching: heat demand divided by the heat pump’s seasonal performance gives the electricity it uses. That is multiplied by each year’s grid carbon factor, which falls as the grid decarbonises, and the carbon from making the heat pump is added.

The lock-in figure is the difference between the two over 12 years. It is roughly what replacing a broken boiler with another one commits a household to.

The breakdown chance comes from an assumed curve, because no UK data on boiler lifetimes is published. It is a Weibull curve with shape 2.5 and scale 14 years, giving an average life of about 12 years. On it, about 11% of 10-year-old boilers fail within a year, 14% at 12 and 19% at 15.

The readiness score adds up the answers to five questions about insulation, radiators, hot water, outdoor space and the fuse board. It is a guide to next steps.

If a radiator keeps its room warm now, it gives out enough heat at the boiler’s water temperature. A heat pump runs cooler water, and a radiator’s output falls with the difference between water and room temperature (to the power 1.3, under BS EN 442). The check works out how much bigger each radiator would need to be at the heat pump’s temperature. A radiator’s length cancels out of that, which is why nothing is measured.

The boiler’s water temperature comes from its heating dial. A 1–6 dial is read as 40–80°C, an unmarked dial as 40–80°C by position, and a dial nobody has changed as 70°C. Water is assumed to cool by 20°C on its way round a boiler system and by 5°C with a heat pump.

For each room, the household says whether the boiler can get it warm on a cold day. A room that overheats has radiator to spare, and one that never gets warm has too little (×0.8 to ×1.25). Rooms kept cool on purpose are not adjusted and are marked untested.

Swaps are compared by heat output per square metre of wall: a Type 22 gives about 1.8 times the output of a Type 11 in the same space. A room within 15% of enough is marked borderline, as that is within the method’s margin of error.

The dial setting is a maximum, and most radiators were fitted with some margin, so the check tends to overstate what’s needed. If the household has kept the house warm with the boiler turned down, that lower temperature is used. If not, the check suggests a week-long test at a lower setting (never below 50°C).