How the engine is verified, and what it assumes
COP is heat out divided by electricity in. Electricity is easy to meter; heat is not. Most customers will never have a heat meter, so the engine estimates heat and must prove how good the estimate is.
Accuracy against heat meters, winter 2026
| System | Brand · model | Metered SCOP | Flow-sensor est. | Error | Temperature-only est. | Error |
|---|---|---|---|---|---|---|
| #116 | Samsung Gen 6 | 4.23 | 4.41 | +4.3% | 3.99 | -5.7% |
| #167 | Nibe F2050 | 3.48 | 3.73 | +7.2% | 3.03 | -12.9% |
| #1 | Mitsubishi Ecodan | 3.53 | – | – | 3.26 | -7.6% |
| #232 | Viessmann Vitocal 150A | 3.75 | 3.82 | +1.9% | 3.77 | +0.5% |
| #295 | Daikin Altherma 3 | 3.34 | 3.67 | +9.9% | 3.66 | +9.6% |
| #2 | Mitsubishi Ecodan | 3.89 | 3.90 | +0.3% | 3.91 | +0.5% |
| #300 | Viessmann Vitocal 150A | 4.21 | 4.26 | +1.2% | 4.00 | -5% |
| #359 ⚑ | Daikin Altherma Split | 1.34 | 2.72 | +103% | 3.34 | +149.3% |
| #382 | Samsung Gen 7 | 4.03 | 4.23 | +5% | 4.04 | +0.2% |
| #441 | Panasonic Aquarea M | 3.84 | 3.96 | +3.1% | 3.02 | -21.4% |
| #51 | Vaillant Arotherm+ | 3.88 | 3.90 | +0.5% | 4.22 | +8.8% |
| #53 | Vaillant Arotherm+ | 4.39 | 4.41 | +0.5% | 4.38 | -0.2% |
| #61 | Nibe F2040 | 3.25 | 3.37 | +3.7% | 3.57 | +9.8% |
| #748 | Grant Aerona HP290 | 4.17 | 4.38 | +5% | 3.87 | -7.2% |
⚑ excluded: the owner reports an unmetered booster heater, so the heat meter is not a fair reference. Window 5 Jan – 4 Feb 2026, hourly means.
The two estimators
Flow-sensor method
heat = flow rate × (flow − return temperature) × 4186 J/kg·K. Needs the unit’s own flow sensor, which most modern heat pumps expose through their cloud API. Flow rates published in m³/h are detected and converted.
Temperature-only method
heat = electricity × η × ideal COP, where ideal COP = (Tflow + 273.15) / (Tflow − Toutside). η = 0.35 was learned leave-one-out from the metered homes above. A production model would learn η per brand and per outdoor-temperature band.
Assumptions and rules
- Electricity price 0.25 GBP/kWh for savings estimates; design conditions −3 °C outside, 20 °C inside.
- A compressor is “running” above 200 W. Space-heating findings need at least 60 heating hours in the window.
- Flow temperature ≥ 50 °C at 0 °C outside is serious, ≥ 45 °C is a warning. Flow − return below 3 K or above 8 K is a warning. Hot water above 30% of electricity is a warning. Cycling ≥ 4 starts/hour is serious, ≥ 2.5 a warning. A unit rated more than twice the fitted heat loss is noted as oversized.
- These thresholds are installer rules of thumb and will be tuned with installer feedback and more winters of data.
- Data: HeatpumpMonitor.org public systems (OpenEnergyMonitor community, CC BY 4.0). Owners published their data; they are not customers.
Sources and licences
- Reference fleet. HeatpumpMonitor.org, public heat-metered systems published by their owners through OpenEnergyMonitor, snapshot of 2026-09-16 (139 systems, 115 with a certified heat meter). Licence CC BY 4.0. We show aggregates and link each system back to its page; values are unchanged apart from aggregation.
- Weather. Weather data by Open-Meteo.com, historical weather API on the ERA5 and ERA5-Land reanalyses, licence CC BY 4.0. Generated using Copernicus Climate Change Service information 2026. The check sends only the postcode’s centre point and the billing period, never personal data.
- Postcodes. Centre points per German postcode from GeoNames, licence CC BY 4.0, held as a local table.
- Typical values. Seasonal efficiency ranges from Fraunhofer ISE field tests of existing houses; heating-demand bands by building age from dena and Verbraucherzentrale; hot-water reference of 12.5 kWh per m² and year from the German building code. Typical values are labelled as such on every check.