Hiding the wiring again: software-defined power, hybrid heating, and the cost, carbon and flexibility of keeping the boiler: evidence from 86 British winters

British policy has written off the hybrid heat pump, which keeps the gas boiler, on modelling that assumed it burned gas only at the peak, while installed hybrids switch on a fixed outdoor temperature or on prices the householder enters. This paper models the control. 32,000 homes in each of 79 British towns, calibrated to 55 measured heat pumps, are simulated half-hourly through all 86 heating years from 1940–41 to 2025–26 at 2025–26 tariffs, under a fixed changeover, a price-aware controller (C-COP) and software-defined power (SD-Power) dispatch, which also holds Britain's demand under a cap. In this design, with Intelligent Octopus Go among the four tariffs, the tariff explains 72% of the variation in whether keeping the boiler lowers running costs (27% without it). Keeping it means keeping the gas standing charge, about £106 a year, so a hybrid is cheaper only where it saves more than that in energy. In the median winter it does so for 13% of homes on a flat tariff, and for 88% on Intelligent Octopus Go, whose daytime rate an all-electric home without pre-heating cannot avoid (61% when both homes pre-heat). On Agile, the one tariff priced from the wholesale market, the answer depends on how historic weather is paired with today's market: over twelve test winters the hybrid wins 11% of homes when each day takes the prices of the same date in 2025–26, and 61% when it takes those of a recent day as cold. On a flat tariff the answer turns on commissioning: over twelve test winters it is 11% under the central assumption, 1% for a well-tuned stock and 36% at the field-trial median. Price-aware control saves £29–£199 a year against a fixed switch at 0 °C; on a flat tariff it emits 411 kg of CO₂ a year more than that switch, and 589 kg more than an all-electric heat pump. On a flat tariff a hybrid stock removes 75% of the winter peak that all-electric heating would add. On dynamic tariffs it herds into the same cheap half-hours and removes 22–61%, in a stress case where homes act together far more than real ones do; letting the herd raise prices, in a simple feedback experiment, shrinks it without removing it. Idealised SD-Power dispatch holds Britain's demand at the flat-tariff hybrid's own peak, so by construction it removes the whole Agile herd above that cap: 22 GW in the worst of 86 winters, or, over twelve test winters with 8,000 homes a town, 23 GW in the main model and 18 GW with heating habits measured in real homes. It costs about £3 a home a year in fuel, before delivery costs. Over twelve test winters, a cap on demand net of wind sends homes to gas less often, for about £1 instead of £3, but lets national demand rise to 70 GW instead of 60. With realistic forecast errors or imperfect scheduling it still takes off 8–12 GW but breaches the cap in every winter; reactive dispatch keeps little. At each winter's own prices since 1961, the hybrid's heating bill was below the all-electric heat pump's in 29 of 65 winters; by construction it is never above gas alone. The results do not make a general case for hybrids: under the central assumptions the all-electric heat pump is the cheaper to run for most homes on a flat tariff, and for almost all where well commissioned. They show that the choice of heat source interacts with heat-pump efficiency, tariff design, carbon and the national peak, and that control must be judged separately from the heating it controls. The licence for home energy controllers should set what a controller may optimise, not only who may hold it. Declaration of interest. The author is a co-founder of LDS Consultancy, which develops the Levant platform and the software-defined power (SD-Power) approach that this paper evaluates. That is a commercial interest in the conclusion that software control adds value. No current bid, tender or funding application relies on this paper. To let readers judge the work independently of it, the model, data preparation, analysis, every number and every source are published openly as a versioned release (Section 4.11), a second implementation checks the model's arithmetic, and the paper reports results that cut against that interest as fully as those that support it. The work received no external funding. LDS Public Class No. 1, version 1.0, 11 November 2026. Web version: https://energy.ldsconsultancy.net/class/ Model, data and results: https://github.com/CFLLDS/gb-hybrid-heat-model (release v1.0.0).