At 11:30 on Saturday 12 September, Britain's transmission-connected wind farms had 20.7 GW of wind available to them. NESO was paying some of the Scottish ones to switch off. Twenty-four hours later the fleet was producing 0.7 GW. That is the biggest 24-hour collapse of wind in the record.

It wasn't a one-off. Since 1 September there have been five collapses of 10 GW or more: on the 1st, 4th, 8th, 10th and 12th. What sets September apart is how deep they went. Three took out more than 70% of installed capacity, and no September since 2018 had even one of those.

The Germans gave us Dunkelflaute for the long, dark calm. There is no word for the sudden fall into one, so here is a proposal: Sturzflaute, a flash lull, by analogy with Sturzflut, a flash flood. A Dunkelflaute is about how long the calm lasts. A Sturzflaute is about how fast you fall into it.

I hadn't intended to write about wind again this month. Last week's piece, Are we about to set a record for curtailment costs?, was about what it costs when there is too much wind in the wrong place. Before that, How Long Is a Dunkelflaute? argued that firmness depends on how long you're prepared to wait. The feedback on both kept circling one question: is this getting worse? Then the wind produced its biggest collapse on record, and it felt sensible to take the question seriously.

Britain's wind fleet is about fifteen years old, and it is having mood swings. The data says three things about its Sturzflauten. They are much bigger. The temperament behind them hasn't changed. And someone has been covering for it.

Read more

On NESO's official numbers, national de-rated margin fell to 4.1 GW — the tightest evening of the summer. Frequency sat below 49.8 Hz for 48 minutes between 18:51 and 20:03, bottoming at 49.656 Hz. At 20:01 NESO requested Emergency Assistance on an inter…

Read more

• 11 April 2026. Wholesale power averaged £4.39/MWh across the day. Essentially free.

• The day cost £124.42/MWh all in: £4.66 wholesale, £27.71 balancing, £92.05 policy.

• Wind resource that day: 83% fleet load factor. One of the windiest days of the period.

• The question the piece answers: how does free power cost £124?

Read more

It has essentially nothing to do with forecasting.

It was, however, the most expensive day of constraint management in at least the five years I have data for — £29.65m — and it beat the previous record, set four days earlier on 4 September, by 11%. Unlike almost every other day in the top ten, it holds that position after adjusting for inflation too.

Updated 9 September 2026. I published the first version of this piece at lunchtime on the day itself, with the afternoon still to come, asking whether a record was coming. It was. This version has the completed day — and corrects two things I got wrong first time round, both flagged in full below. Thanks to Robin Hawkes for catching the second one.

The forecast was right

The wind forecast Elexon publishes is an unconstrained forecast: what the fleet would generate if the grid let it. The outturn it is plotted against is metered output, which is what the fleet generated after the control room finished intervening. The difference between those two things is not error. It is instruction.

I reconstructed the day's instructions from Elexon's physical notification and bid-offer acceptance data across all 232 wind BM units. Averaged over the full 24 hours:

• Day-ahead forecast: 16,997 MW

• What the wind farms themselves declared they intended to generate: 17,912 MW

• Metered output: 13,302 MW

• Volume NESO instructed off: 4,767 MW

• Residual, once you add the curtailed volume back: +1,071 MW

Curtailment accounts for 129% of the apparent gap. Add back what was switched off and the fleet comfortably beat the day-ahead forecast. More tellingly, the wind farms' own notifications came in 914 MW above NESO's forecast — the operators who get paid to know agreed the wind was there.

As a forecast miss, the day was unremarkable: 65th worst out of 2,071. On curtailment it ranked 8th. On residual error — the part that is actually forecasting — it ranked 1,865th, which is to say the forecast was better than usual. That spread between the three rankings is the whole story.

This is not a quirk of one day. Across five years the average gap has widened from 203 MW in 2021 to 1,319 MW in 2026, and curtailment explains between 72% and 124% of it depending on the year — over 90% of it in 2021, 2022, 2025 and 2026. The residual error has stayed flat at a standard deviation of about 1.1-1.4 GW while average output nearly doubled from 5.8 GW to 9.4 GW. In relative terms the forecast has quietly got better the whole time.

The chart that looks like deteriorating forecast skill is a chart of the constraint problem.

One in, one out

Here is the same day from the other side. When wind comes off in Scotland, something in England has to come on.

Across the full day, gas was instructed up 4,724 MW while wind was instructed down 4,767 MW. In energy terms: 113 GWh of gas turned on against 114 GWh of wind turned off. A difference of under 1%.

The striking part is where gas started. The CCGT fleet's own traded position for the day was 748 MW — with wind this strong, the wholesale market had barely bought any gas at all. NESO ran it at 5,400 MW. Gas generated more than seven times its market position, almost entirely on instruction.

Three checks say this reconstruction is sound. Everything nets to within a few tens of MW across all fuels, as energy balance requires. The post-instruction levels match metered outturn to within 1-2% for gas and wind. And Elexon's settlement stack — an entirely separate dataset, and the one Robin Hawkes' excellent Wasted Wind tracker uses — independently gives 119 GWh of wind bid down against 117 GWh of gas offered up.

Five gigawatts is not a rounding error

It is worth stopping on that number, because "4.8 GW of curtailment" slides past the eye far too easily.

Across Tuesday Britain switched off more wind than its entire nuclear fleet was generating — 4.77 GW discarded against 3.25 GW of nuclear output. It was 1.8 times the biomass fleet. It was 88% of everything every gas station in the country was producing. Had it been allowed to run, that single block of wind would have been the second largest source of electricity on the system, all day.

In total, 8 September curtailed 114 GWh — the eighth biggest day in the five-year record, and 20% of everything Britain consumed. One electron in five, generated and then thrown away.

Last Friday was bigger still. 4 September curtailed 127 GWh, a continuous 5,303 MW for twenty-four hours and 23% of national demand — the third biggest day on record. If you prefer it in households: at the standard 2,700 kWh a year, Tuesday's 114 GWh is a full day's electricity for 15.4 million homes, about 55% of every household in the country.

Why now? The wires are still on their summer rating

Britain's boundary limits are set on static seasonal ratings, not dynamic line ratings. They step up on the calendar, in autumn, when conductors can be run harder in cold air. They do not respond to the weather actually happening.

Which is the problem, because the weather arrived early.

Across the first eight days of September, B6 — the Scotland-to-England boundary — has been running a mean limit of 4,120 MW against a winter mean of 5,272 MW. It is still 22% below its cold-weather capability. The day-ahead flow the market wants to push across it is 5,298 MW. Utilisation of 127%, binding in 63% of half-hours.

B4/B5 in central Scotland is worse. Its September limit of 1,867 MW is essentially unchanged from the summer rating of 1,832 MW, against a winter mean of 2,901 MW. Flow wants to be 2,990 MW. Utilisation of 162%.

So: winter-strength wind, summer-strength wires. Every megawatt of that mismatch has to be resolved by switching wind off in Scotland and paying something in England to make up the difference.

And replacement power has got much more expensive

This is the part that has changed fastest, and it is not really about volumes at all.

The price NESO pays gas for constraint turn-up has gone from £100/MWh in December 2025 to £217/MWh in September 2026 — a 2.2x rise in nine months, and up every single month since February.

Some of that is simply the market. The short-term wholesale reference price — Elexon's market index price, the within-day benchmark settlement uses, and the right counterfactual for actions taken this close to real time — rose over the same period too, from £74.70/MWh to £111.00/MWh. That is a real and honest chunk of the answer, and it would be wrong to present the whole rise as anything else.

But it does not cover it. The premium over wholesale has widened from around £25-45/MWh through most of 2025 to £106/MWh now. When the system needs specific machines in specific places, it is not buying at the market price — it is buying from whoever is left, and increasingly from a smaller and dearer set of them.

The record, and where it came from

The completed bill for 8 September, from Elexon's settlement cashflows: £29.65m. Of that, £3.05m went to wind farms to stop and £26.60m to buying the electricity back.

A word on how that second number is arrived at, because it is the difference between a defensible figure and an inflated one. The control room turned down 144 GWh of generation on Tuesday, of which 119 GWh — 82.7% — was wind. It turned up 145 GWh to replace it. Charging all of that turn-up to wind curtailment would overstate the case, because some of it was replacing the other 17%. So the turn-up cash is attributed by wind's share of bid-down volume. Gross, the day's balancing turn-up came to £35.22m; wind's attributable share is £29.65m.

That convention matters for comparison as well as for honesty: it is the one Wasted Wind uses. On this basis my total for 1-8 September is £130.2m against Wasted Wind's £128.8m — agreement to within 1.1%, from two independent reconstructions.

The previous record, 4 September 2026 at £26.67m, was four days old when it fell.

The most interesting thing about the new record is how it was set. Compare the two days:

• Wind switched off — 4 Sep: 123 GWh · 8 Sep: 119 GWh

• Paid to wind to stop — 4 Sep: £4.95m (£40/MWh) · 8 Sep: £3.05m (£26/MWh)

• Gas turned up — 4 Sep: 91 GWh · 8 Sep: 117 GWh

• Gas price for replacement — 4 Sep: £207/MWh · 8 Sep: £231/MWh

• Attributed replacement cost — 4 Sep: £21.72m · 8 Sep: £26.60m

• Total — 4 Sep: £26.67m · 8 Sep: £29.65m

Tuesday switched off less wind than Friday, and paid the wind farms less per megawatt-hour to do it — £26/MWh against £40/MWh. The record was set entirely on the replacement side: 29% more gas, bought at 12% higher prices. The bid side of the ledger got cheaper and the bill still went up 11%.

That matters, because the public conversation about curtailment is almost entirely about the money paid to wind farms to switch off. On Tuesday that was £3.05m of a £29.65m day — barely a tenth of it. The other 90% is what it cost to put the electricity back.

There is a second detail worth having. The morning cost £17.7m of it and the remaining ten hours £11.9m. The record was set on a decelerating day: the evening was cheaper per hour than the morning, and it broke the record anyway.

But what about 2022?

This was the first objection I had to answer, and it is a good one. In autumn 2022 gas was at war prices and the turn-up rate NESO paid ran at £350/MWh in March 2022 against £197/MWh this month. A far smaller volume could have cost far more.

So I costed every plausible crisis-era day directly from the settlement stack — 190 complete days in all, chosen as the top days by curtailment volume and by gas turn-up across the whole 2021-2026 record — and restated everything in 2026 prices using ONS CPI.

It changes the table, and it does not change the answer — but it makes it a much closer thing than I first wrote. In today's money the earlier era reclaims most of the top ten: 18 November 2021 (£28.66m real, £22.97m at the time) and 6 October 2022 (£27.94m real, £24.67m at the time) take second and third, and only three of the top ten are 2026 days. 8 September 2026 still leads — but by 3.4%, not by the double digits I claimed in the first version. That is inside the range where someone else's reasonable attribution choice could reorder the top two, and it should be read as "as expensive as the worst of the energy crisis" rather than "comfortably worse".

Two corrections. I want to be straight about both, because the first version of this piece got them wrong and they both cut the same way — against my own argument.

The first. That version said eight of the ten most expensive days on record were 2026 days. They are not. Elexon does not populate the finalPrice field on pre-2023 settlement records — it returns zeros rather than nulls — and taking those at face value silently under-costs the entire crisis era. Fall back to originalPrice for those days and pre-2023 costs rise by 20-45%, putting most of the top ten back in 2021-22.

The second, and the bigger one. The first version quoted gross balancing turn-up: wind's bid cost plus all the turn-up cash on the day. That charges wind curtailment for replacing generation that wasn't wind. Robin Hawkes, who runs Wasted Wind, pointed out that his numbers came in materially below mine, and he was right to. Attributing turn-up by wind's share of bid-down volume — the method my own data note claimed I was using, and the one his tracker uses — takes 8 September from £35.22m to £29.65m, and 4 September from £30.97m to £26.67m. Every figure in this piece is now on the attributed basis. The record survives the change; the margin over 2021-22 does not survive intact.

The corrected picture is less flattering to my argument and more interesting: 2026 is not burying the crisis era across the board. It has simply produced the two single worst days, and it has done it without crisis-era gas prices.

On cost per megawatt-hour of national demand, Tuesday comes to £51/MWh — high, but not the highest: 11 June 2022 reached £58/MWh on a much smaller demand base. So the honest claim is the cash one: the biggest single-day constraint bill on record, in nominal and real terms, and not the highest intensity ever recorded.

An important caveat on any of these £/MWh figures: nobody is billed like this. Since April 2023 BSUoS has been a fixed, forward-looking tariff, set six months ahead and charged wholly to demand. A day like Tuesday does not land on Tuesday's consumers — it is smeared across the tariff period and, in practice, across everyone. The figure is a way of sizing the cost against the electricity actually being consumed, not an invoice anybody received. It is also gross: the system would have paid something for that energy regardless, so it overstates the true incremental cost.

And there is a second reason to hold it loosely: the denominator is unusually small. These have been mild, bright, breezy days. Britain's daily demand has fallen from 28.3 GW in 2021 to 25.4 GW in 2026, so any per-MWh comparison flatters the present.

The two eras got there by opposite routes. In 2022 the system paid crisis prices for small volumes: 30 September 2022 involved 42 GWh of wind switched off at £277/MWh on the replacement side. In 2026 it pays merely high prices for enormous volumes: Tuesday was 119 GWh switched off and 117 GWh bought back at £231/MWh. Three times the wind, at four-fifths the price.

There is a lovely detail buried in the 2022 numbers, too. On the worst crisis days the bid side was a credit, not a cost — on 30 September 2022 wind farms paid the system to be switched off, because with wholesale prices that high they were better off stopping. That does not happen any more. In 2026 we pay them to stop, and then pay gas seven to nine times as much to start.

The volumes alone were never going to do this

Tuesday ranks 8th of 2,067 days for curtailment volume. 4 September, at 127 GWh, was the third biggest. Big — but neither is unprecedented, and this is the point. Days averaging over 4 GW of curtailment happened 8 times in 2024, 20 times in 2025, and 12 times so far in 2026. On volume, this year is if anything running slightly behind last year's pace.

It is the price that has moved. Same problem, roughly twice the cost per megawatt-hour to fix it.

The wider picture is the same story. September has curtailed 580 GWh in eight days, well over August's 406 GWh for the whole month, at a cost of £130m across those eight days alone — against Wasted Wind's £128.8m for the same period, which is as close as two independent reconstructions of this get. Wasted Wind puts 2026 at over 7 TWh and £1.18bn so far — a run-rate of about 10.5 TWh against 9.86 TWh for all of 2025.

What this actually is

There is a version of this story that says the wind forecast is getting worse. The data says the opposite.

There is a version that says we are wasting more wind because we are building more wind. Partly true, but 2026's volumes are running level with 2025's — while the cost has roughly doubled.

What is actually happening is narrower and more fixable. We have built generation in Scotland faster than we have built wires out of it, we rate those wires on a seasonal calendar rather than on conditions, and when the gap opens we close it by buying replacement energy in a market with very few sellers. The first problem takes a decade. The second is a ratings and operational question. The third is a market design question — and it is the one moving fastest in the wrong direction.

Two records in five days, both set by a grid doing exactly the same thing, and 90% of the bill was not what we paid to waste the wind. It was what we paid to replace it.

A note on the data, for anyone reproducing this. The wind actuals on Elexon's day-ahead wind-and-solar page (dataset AGWS) are badly under-reported before 2025 — offshore averages 3.0 GW across 2021-24 against 5.8 GW in 2025, and on 21 October 2024 it reports 4.9 GW of wind when metered output averaged 11.8 GW. Building a historic comparison on those actuals produces "forecast errors" of −73% that are pure artefact. Use FUELHH for metered wind instead. The solar series on that page is fine.

Separately, and more consequentially for any cost comparison across eras: on settlement stack records before 2023, finalPrice is returned as zero rather than null. Cost anything pre-2023 on that field and you will understate the energy crisis by 20-45% and conclude, wrongly, that recent days dominate the all-time table. Fall back to originalPrice whenever finalPrice is missing or exactly zero.

Curtailment volumes here are reconstructed from physical notifications versus post-acceptance levels across all wind BM units, and cross-check to within 2-3% of Wasted Wind's settlement-based figures for 2024, 2025 and 2026. Constraint cost attribution follows the approach used by my own turn-up tracker: wind's bid cash in full, plus turn-up cashflows multiplied by wind's share of that day's total bid-down volume (82.7% on 8 September). The first version of this piece stated that method but did not apply it — it reported gross turn-up instead, which is 14-16% higher on these days. Corrected throughout.

Wind and Solar Power

The biggest growing source of power over the last 15 years, wind and solar power are both totally dependent on short-term weather. They are not firm in a traditional sense. You can't bank on them being able to generate beyond the forecastable horizon of the weather.

With the UK's geography and planning system, wind also tends to be built further north, creating an additional dependency on the successful expansion of grid transmission capacity. If it's "too" windy in Scotland relative to the rest of Britain and demand, some of the wind power generated will have to be curtailed and substituted for elsewhere.

Wind and solar do have some positive attributes from a reliability perspective. They are inherently modular, comprising individual turbines (up to 10MW) and arrays of panels. Compared to a typical thermal power plant unit of 500MW, they can often operate without onsite staff, and individual turbines/arrays can fail without causing a wider outage.

While bad weather can severely limit wind and solar generation, at certain times it is possible to be much more confident of a minimal generation output. For example, midday solar generation even on an overcast or rainy day will hit a certain minimal level, even if this is just a fraction of the potential. The diversity of wind farm locations similarly means that there is always some wind generation somewhere. And if you consider wind and solar as part of a future energy system with more storage and flexibility, what becomes more relevant for firmness is the rolling average of the combined national wind/solar portfolio over time.

Figure 1 — firmness is a function of the averaging window

Over three complete years the combined GB wind and solar fleet averaged 11.2 GW. Its worst half hour was 0.2 GW — 2% of that. Its worst seven days was 3.3 GW, 29%. Its worst thirty days was 6.7 GW, 60%. Nothing about the fleet changed between those numbers; only how long you are allowed to wait.

The right-hand panel asks the same question one season at a time, each normalised to its own mean, and the shapes differ in a way that names which technology is doing the work:

• 30 minutes — winter: 2% · spring: 5% · summer: 4% · autumn: 4%

• 1 day — winter: 10% · spring: 28% · summer: 33% · autumn: 15%

• 7 days — winter: 44% · spring: 51% · summer: 57% · autumn: 29%

• 30 days — winter: 76% · spring: 72% · summer: 77% · autumn: 68%

• season mean — winter: 13.0 GW · spring: 10.5 GW · summer: 10.0 GW · autumn: 11.3 GW

Solar buys firmness by the day; wind buys it by the month. Summer has the lowest mean output of the four seasons and the highest one-day floor — a third of its average, because the sun turns up every single day — but that daily floor buys nothing extra at a week or a month, so the curve flattens. Winter is the mirror image: nothing is guaranteed daily, so its one-day floor is 10% — summer's daily floor is three and a half times winter's — but no winter month in three years has been becalmed and its thirty-day floor is the best of the four at 76%. Autumn is the worst place to be at a week, 29%, which is the October 2025 wind drought that figure 7 also picks out. By thirty days the seasons have converged: all four sit inside 68–77%.

Figure 2 — solar fails on a timetable, wind fails on the weather

That difference in duration has a mechanism, and it is not that winter is windier. It is that the two technologies fail on different clocks.

Solar has a guaranteed daily zero. Every summer night output falls back to the wind alone, and it does so on schedule — 6.7 GW at 04:00 against 15.3 GW at half past twelve. Wind has no daily anything. Average winter output at 04:30 and at noon differ by 6%: the clock tells you almost nothing about how windy it is going to be.

Put as a variance decomposition — how much of each season's spread is explained by its own average daily profile:

• winter — the clock explains: 4% · wind swings across the day: 6% · solar above 1 GW: 6 h a day

• summer — the clock explains: 36% · wind swings across the day: 20% · solar above 1 GW: 12.5 h a day

Which is why the shortfalls have different shapes. Summer's are frequent, regular and short — nine in ten are over inside 21 hours, and the worst in three years ran 3.9 days. Winter's are rarer but open-ended: nine in ten inside 2.4 days, worst 6.6. Autumn is worse than either at 10.4 days — the October 2025 drought again.

Storage that empties and refills every night is a different asset from storage that has to ride out a week, and the same fleet needs both, in different seasons.

One check worth stating, because these are metered series and metering is net of curtailment — which is itself diurnal. Against the transmission fleet's own 2025 availability:

• winter — delivered swing: 7.8% · available swing: 3.5% · curtailment peaks: 06:00, 1.5 GW

• summer — delivered swing: 17.0% · available swing: 17.7% · curtailment peaks: 19:00, 1.3 GW

Winter's small daily shape is mostly an artefact — the network bidding wind off overnight when demand is lowest. The resource underneath is flatter still. Summer's is weather: delivered and available swing by the same amount and peak in the same late afternoon, which is the ordinary onshore diurnal cycle as daytime heating mixes momentum down. So the 4% figure overstates how much the winter clock tells you, and the 36% is real.

Figure 3 — solar's floor is real, and it is a summer floor

Across three Junes, the worst midday half hour GB solar ever produced was 3.6 GW — 41% of the median June midday. Whatever the weather, the middle of a June day is worth something. December's floor is 0.3 GW, 14% of an already small median.

Figure 4 — portfolio diversification buys a great deal, but not a floor

Portfolio diversification cannot add energy, so it can only move it — and the left panel is that trade. The pooled curve runs below the median farm for the first 28% of the year and above it for the other 72%. The fleet is never all-out at once, which costs it the top: its best half hour is 90% of its parts' summed peak, while every individual farm reaches 100% of its own. It is almost never all-off at once, which buys it the bottom: at the 90th percentile of the year the median farm is producing nothing and the fleet still has 10% of its peak.

The right panel counts the same fact differently: the median single farm sits below 5% of its own peak for 24% of the year, the pooled fleet for 3%. But at 09:00 on 22 January 2025 the whole fleet fell to 0.1% of its peak, out of 23.6 GW. Britain is not wide enough to be its own portfolio.

Output here is availability, not metered — a farm that was paid to stop is not a farm with no wind. On metered output the median farm reads 30% rather than 24%, and Moray East, half of whose year was bid off, looks becalmed 39% of the time against a true 23%. Curtailment is a separate argument and gets the next figure.

Figure 5 — in a national calm, what matters is the distance, not the site

If diversification is what buys the floor, which parts of the portfolio are actually doing the diversifying? Take the 10% of 2025 when the pooled fleet was at its lowest — the fleet averaged 7.0% of peak against 40.4% over the year — and ask what each unit was doing.

Start with the thing that is true of all of them. In a national calm the fleet keeps 17% of its own normal output, and there is no farm in Britain that keeps more than 36%. So nothing in the table below holds up in any absolute sense: the entire spread runs from losing two thirds of your normal output to losing nine tenths of it. The question is only which farms are still contributing something when the rest have effectively stopped.

On that, it is not simply the windy sites. Annual load factor explains only a third of how a farm performs in a calm. The rest is geography, and the pattern is the extremities of the map against the middle:

• Viking, Shetland — annual: 44.2% · in a calm: 15.8% · keeps: 36% · · worst in a calm: Harestanes, Dumfriesshire · annual: 22.6% · in a calm: 2.1% · keeps: 9%

• Rampion, off Sussex — annual: 41.9% · in a calm: 14.2% · keeps: 34% · · worst in a calm: Griffin, Perthshire · annual: 21.3% · in a calm: 2.5% · keeps: 12%

• Neart na Gaoithe, off Fife — annual: 41.3% · in a calm: 11.7% · keeps: 28% · · worst in a calm: Whitelee, south of Glasgow · annual: 23.7% · in a calm: 2.9% · keeps: 12%

• Galloper, off Suffolk — annual: 48.6% · in a calm: 11.4% · keeps: 23% · · worst in a calm: Black Law, Lanarkshire · annual: 21.3% · in a calm: 3.1% · keeps: 15%

• East Anglia One, off Suffolk — annual: 48.8% · in a calm: 9.7% · keeps: 20% · · worst in a calm: Crystal Rig II, Lammermuirs · annual: 26.6% · in a calm: 3.4% · keeps: 13%

Read the right-hand block as a map and it is one place: the Southern Uplands and the central belt, every one of them within about 80 miles of Glasgow. The left-hand block is Shetland, Sussex, Fife and Suffolk — the corners.

And 15.8% is not a good outcome, which is the other half of reading that table honestly. Viking is the best-sited wind farm in Britain for exactly this problem and in a calm it runs at about a seventh of its nameplate. Take the counterfactual seriously — rebuild every farm in the country to Viking's standard, ignore that its site is unrepeatable — and the units this figure measures — those above 100 MW, 17.2 GW of the fleet's 23.6 — go from 1.3 GW in the median calm half hour to 2.7 GW, against a winter peak near 45 GW. The prize for getting the geography perfectly right is worth about one and a half gigawatts. It is not nothing, and it is not a floor.

And the corners are the expensive end of the fleet: four of the best five are offshore and all five of the worst are onshore Scotland, with AR7 clearing offshore wind at £91.20/MWh against onshore at £72.24, before transmission charges that run the same way. Whether a premium like that is repaid by about 1.5 GW of avoided firm capacity is not obvious — the less so because, as a later part will show, you need very nearly as much gas standing behind either fleet.

Moray East is the case that makes the point, because it is the one that cannot be explained by its site being poor. It has one of the best annual load factors in the country at 44.0%, and in a national calm it delivers 6.9% — it keeps 16% of its normal output where Viking, on a near-identical annual figure of 44.2%, keeps 36%. Beatrice, its neighbour in the Moray Firth, does the same: 41.3% across the year, 6.7% in a calm.

What separates them from Viking is not the turbine, and it is not how much wind the site gets — the annual means are within two tenths of a point of each other. It is how that wind is distributed. Shetland sits 250 km further north, which is far enough to fall outside the blocking high that becalms the mainland: same average, different tail. And the tail is what a firm system is sized against.

And the worst retention in the country belongs to good sites, not bad ones. Race Bank off Norfolk averages 42.6% across the year and keeps 12% of it in a calm; West of Duddon Sands 44.1% and 13%; Triton Knoll 39.0% and 12%; Ormonde 34.8% and 11%. A farm can be among the windiest in Britain on an average day and among the least useful on the day it is needed.

So there is a real effect here, and it is worth being precise about what it is — because the obvious reading, that Britain should buy the corners and stop building in Lanarkshire, is only about half right.

The corners do decorrelate. Across 2025 the average pair of British wind farms moved together at +0.48. In the calm decile that falls to +0.09, and Viking and Rampion — Shetland and Sussex, just over 1,000 km apart, about as far as two British wind farms can get from each other — go from +0.07 across the year to −0.14 in a calm. Textbook diversification: when the fleet is struggling, the units stop moving as one.

And it buys almost nothing. The fleet still fell to 0.11% of peak at 09:00 on 22 January. In the median calm half hour 78% of the fleet is below a tenth of its peak at the same moment, and in 79% of them there is not a single farm anywhere in the country above half of its. In the worst half hour of the year, the best-performing wind farm in Britain — out of forty-five, spread over a thousand kilometres — was managing 4.2%.

Which is the resolution of the paradox, and the thing the geography is really telling us. The farms genuinely do become independent of each other in a calm. But independence at 7% of peak is independence between small numbers, and small numbers that are uncorrelated still sum to a small number. What a firm system needs is not for the farms to stop moving together; it is for somebody to be windy — and in a British calm nobody is, because the high-pressure system that causes it is bigger than the country. Britain is roughly a thousand kilometres from end to end, and that is not far enough to hold two weathers.

So the marginal value of a wind farm to a firm system is indeed how far it sits from everything already built, and Britain has been buying the wrong thing on that measure — the cheapest sites are in the middle of the existing cluster. But the ceiling on what the right answer buys you is low, and it is set by the size of the island. Which is why the search for a floor has to leave it, and why figure 16 goes looking for the wind on a map of a continent rather than a country.

Figure 6 — the constraint binds in surplus, not in scarcity

Britain threw away 10.2 TWh of available wind in 2025, 13% of everything the transmission-connected fleet could have produced. That sounds like it should make the firmness problem worse. It does not, and the reason is the shape of when: curtailment correlates +0.75 with fleet output and −0.40 with residual demand. Below 10% of peak output the curtailment rate is 0.3%; above 80% it is 24%.

Against system tightness the same point from the other side. The tightest tenth of half hours carries 3.4% of the year's curtailment, the tightest hundredth 0.1%, and in the single tightest half hour of 2025 — 8 January, 44.4 GW of residual demand — just 0.22 GW was bid off.

It is not never, though. 147 hours of 2025 had both top-decile tightness and more than a gigawatt curtailed, the worst 4.3 GW on 3 February. That tail is the thing that grows if Scottish wind is built out ahead of the boundary. And it is a boundary problem, not a fleet-wide one: the top ten units carry 52% of all curtailment and the top twenty 74%, with Seagreen and Moray East and West the worst six — all north of B6.

Figure 7 — even at CP2030 capacity, these weeks need 37–45 GW from somewhere else

Ranking Dunkelflaute by lowest output finds the wrong weeks. The quietest seven days of these three years were in October 2025 — 3.3 GW of wind and solar — but demand was a mild 32 GW, so the gap everything else had to fill was 29.0 GW, barely above the ordinary 20.9 GW. Rank instead on the gap itself and every one of the four worst weeks lands in the heating season, the worst being 16–23 January 2025: 5.7 GW of wind and solar against 38.4 GW of demand, a 32.7 GW gap held for seven days, peaking at 43.5 GW. The failure mode is not a dip, it is a plateau.

The temperatures confirm it. All four weeks run 11–14 heating degree days a day against a three-year average of 4.9 — colder than 91–97% of all weeks. The October week with the lowest output sits at 4.1, dead average. Low wind is common; low wind in a cold snap is what costs money.

Scaling the same weather to Clean Power 2030 capacity — wind ×2.3, solar ×2.2 — lifts that week from 5.7 to 13.1 GW, while CP2030's demand growth lifts the other line from 38.4 to 41.2 GW. The average gap narrows from 32.7 to 28.0 GW, and a doubled fleet still covers only 32% of that week's demand against 15% today.

The worst half hour barely moves at all, and that is the number capacity adequacy actually turns on:

• 21–28 Jan 2023 — peak gap today: 40.7 GW · peak gap at CP2030: 37.7 GW

• 27 Feb – 6 Mar 2023 — peak gap today: 37.7 GW · peak gap at CP2030: 36.9 GW

• 26 Nov – 3 Dec 2023 — peak gap today: 44.2 GW · peak gap at CP2030: 45.3 GW

• 16–23 Jan 2025 — peak gap today: 43.5 GW · peak gap at CP2030: 45.1 GW

In two of the four weeks the peak gap gets worse. Doubling a fleet that is producing almost nothing still produces almost nothing, while the demand uplift is there regardless. Seven days is still seven days: what closes the remainder is storage, interconnection and flexibility, none of which is netted off above.

Figure 16 — the interconnection half of that sentence

Every proposed answer to a British wind drought is a line on a map to somewhere else: the Norwegian interconnectors, the North Sea grid, the Baltic, Xlinks' 3,800 km cable from Morocco. So it is worth asking the same question of a box running from the Sahara to the Barents Sea that figure 14 asked of Britain's own — where was the wind, in the weeks Britain ran short?

Britain averaged 7.34 m/s across those four weeks against 11.16 normally — 66% of its own average. All nine regions, with what each was doing relative to its own normal, and how far it is from the middle of Britain:

• Norwegian Sea — mean: 12.01 m/s · of its own normal: 102% · distance: 1,326 km

• The Faroes and the Iceland gap — mean: 11.29 · of its own normal: 83% · distance: 1,026 km

• Iceland — mean: 9.50 · of its own normal: 92% · distance: 1,480 km

• Gulf of Lion to Sardinia — mean: 8.45 · of its own normal: 107% · distance: 1,627 km

• The Baltic — mean: 7.81 · of its own normal: 95% · distance: 1,476 km

• Bay of Biscay — mean: 7.60 · of its own normal: 73% · distance: 962 km

• The eastern North Sea — mean: 7.54 · of its own normal: 67% · distance: 598 km

• Great Britain — mean: 7.34 · of its own normal: 66% · distance: —

• Morocco (Xlinks) — mean: 6.24 · of its own normal: 94% · distance: 2,844 km

• Iberia — mean: 5.22 · of its own normal: 87% · distance: 1,628 km

Seven of the nine beat Britain, so the first thing to say is that the wind was not switched off across a continent. But read the middle column. The nearest ground that beat Britain was having almost exactly Britain's bad week — the eastern North Sea at 67% of its own normal against Britain's 66%, 598 km away, which is to say inside the same weather system. Biscay, the next nearest, was at 73%. Everything that was genuinely windy for itself was 1,000 km or more distant: the Norwegian Sea at 102% and the Gulf of Lion at 107% were the only two regions of the nine running above their own average, and they sit at opposite ends of the map, 1,326 and 1,627 km away in almost opposite directions.

Morocco sits second from bottom there, though judging Xlinks by the wind column misses that it was mostly a solar project — 247 W/m² against Britain's 59 over those same weeks, and still at 381 W/m² at 5pm when Britain's residual demand peaks — and in any case DESNZ turned it down in June 2025.

So the escape from a British wind drought exists, and it is a long way off in two different directions. That is a statement about how much cable, not about whether the wind is there.

Worth noting what CP2030's extra demand actually is. Total GB demand grows 263 → 287 TWh, +11%, but residential demand falls: around 20 TWh of lighting and appliance efficiency more than outweighs the heat pumps added, since the plan only reaches the 600,000-installations-a-year rate by 2028. The growth is data centres, from 5 to 22 TWh, and hydrogen electrolysis at 11 TWh. On CP2030's own numbers the thing driving electricity demand growth this decade is not heat or transport — it is AI and hydrogen.

What this measures, and what it does not

Every gap in this piece is a gap against demand as it is now. The four worst weeks were scored against a Britain that still heats itself by burning things, in buildings the electricity grid has never had to warm. That is the quiet assumption underneath the whole exercise, and it is the one about to stop being true.

And it measures the problem, not the answer. There is no storage in any of these numbers, no demand-side flexibility, no interconnector flows — a wind farm at 4% of its peak reads as 4%, however full the batteries happen to be. That is the right way round for a first post, because every one of those things is bought in proportion to the hole it fills, and the hole is what figure 1 through figure 16 have been sizing. But it does mean nothing here shows that storage and flexibility cannot do the job. It shows what they would have to do: carry a fleet that spends the worst week of the year at 29% of its own average, in the season when demand is highest, with no help available from anywhere inside the island. Later parts put numbers on that — what within-day flexibility is worth on its own (about 11 GW of the apparent problem, it turns out), how much seasonal storage a system like this needs, and what it costs to keep the plant that covers the rest.

It matters because of what a data centre is and a house is not. A data centre is 22 TWh arriving as a flat band — more energy, same shape, and a shape this system is already built for. Heat is not flat. It is the most seasonal thing Britain does, it is concentrated into the same months as the calms, and it is currently carried by a network almost nobody thinks about: on a January morning the gas grid delivers close to three times what the electricity grid has ever delivered at its own all-time peak.

That network is scheduled to be dismantled and its job handed to the wires measured above — in exactly the weeks measured above. So the question the next part asks is not whether heat can be electrified. It is what happens to the shape of demand when it is, and whether the thing doing the heating still works when it is coldest.

Next, part 2: The Heat Grid Is the Big One — why the network being retired is larger than the one taking over, and what a heat pump actually manages on the coldest morning of the year.

One thing to be clear about from the start: this is a story about houses. Britain's larger business sites have been settled on their actual half-hourly consumption for decades, and mid-sized ones since 2017. Offices, factories and shops have had a market that knows what they really did for as long as most of their meters have existed. It was homes — and only homes — that got the average.

Act 1 — the divergence is real, and nobody looks at it

Wednesday 18 June 2025 and Wednesday 15 October 2025. Each line is normalised to the supplier's own day, so this is shape, not size — six companies selling identical electrons off the same wires to visibly different customers.

Elexon's P114 settlement files carry supplier × GSP region × half-hour metered volume for every supplier in GB, going back decades, free with a portal key. Almost nobody reads them.

On a sunny June weekday Octopus and OVO's midday volume collapses to about 1.05% of their own day per half hour. British Gas and EDF sit near 2.1%. By October all six lines lie on top of each other.

Ooh arrays. The same June Wednesday split fourteen ways, October dotted behind. The dip tracks irradiance and roof stock as you move north: about 1.5% of daily volume per half hour in South Scotland, 0.5% in Southern, below zero in the South West — and nothing at all in London.

In the South West, Octopus goes net negative at midday: its customers there export more than they import. In London there is no dip at all, for anyone. Flats have no roofs. That contrast is what proves this is solar and not tariff behaviour — a behavioural effect would show up in London too.

Metered domestic export sat at four to six thousand MPANs for five years, then went near-vertical from mid-2023. Octopus is roughly 61% of the 2026 stock and almost all of the growth.

The meter counts confirm it from a completely different direction: 39.4 domestic export MPANs per 1,000 domestic meters at Octopus, against 4.0 at EDF — and Octopus holds about 61% of every metered domestic export MPAN in Britain. One caveat to carry: under the Feed-in Tariff export was deemed, not metered, so part of that curve is existing solar homes being moved onto metered export rather than new panels. Octopus built the tariff that made registering worth doing, and took the registrations.

One distinction to hold throughout, because it trips people up. When this piece calls a book domestic or business, it means by volume, not by customer count. EDF is 24% domestic by volume — but 80% of its meters are households. Its business customers are simply enormous: a few thousand industrial sites outweigh two and a half million homes. British Gas is 56% domestic by volume and 83% by meters; E.ON 50% and 92%. Only at the pure ends do the two measures agree — Octopus 87% and 94%, Utilita 96% and 99%. Load shape is a volume-weighted thing, so volume is the right denominator here; but "EDF is a business supplier" means its energy is, not its customers.

The books differ on more than solar. Peak time runs as a clean scale — SmartestEnergy 09:00, Drax 10:30, EDF 16:30, British Gas 17:00, Octopus 18:00 — and Drax's evening share is 20.77% against a flat book's 20.83%, which is to say dead flat. The regional footprints are still frozen at privatisation: ScottishPower indexes 294 in South Scotland and 214 in Manweb, OVO 541 in North Scotland (the Scottish Hydro book it bought from SSE in 2020), EDF 151 in London. Octopus runs 68 to 132 everywhere — the only genuinely national book in GB, because it never inherited a monopoly.

Act 2 — except most of that was a fiction

Under legacy settlement, non-half-hourly meters were read — but only for the total. A meter reading, smart or otherwise, became an annual quantity of electricity, and settlement then spread that quantity across the 17,520 half hours of the year using coefficients from one of eight Profile Classes — national curves produced centrally from a few thousand sample meters. The arithmetic ran the wrong way round: the sum was measured and the shape was supplied. Every supplier in a region got the same shape.

Which is what makes the smart meter rollout such an odd artefact. A smart meter records 48 values a day. Settlement used one of them — the running total — and then told itself what the other 47 must have been. For fourteen years the distinguishing feature of a smart meter was, for this purpose, discarded on arrival.

Two numbers do the work here.

Utilita and Utility Warehouse — prepayment against multi-service bundle, about as different as two customer bases get — have settled shapes differing by a mean of 0.09 percentage points across all 48 periods.

And Utilita has one seventeenth of EDF's solar customers but twice the settled-shape solar response, because the profile classes carry average solar. The shape was mostly measuring how domestic you were.

Worth admitting in the piece: my own first reading of Octopus's fat overnight was that it was Economy 7 profile-class share. It wasn't — the step is at 23:30 in all fourteen regions, and Economy 7 windows are set regionally. It is Intelligent Octopus Go. But that is the mistake most readers will make, and making it in public is how you show the trap.

Act 3 — the fiction is being removed, live

Two July Wednesdays, one year apart. In 2025 all three are smooth curves. In 2026, with about 70% of GB meters settled half-hourly, British Gas and EDF are still smooth and Octopus is not.

Market-wide Half Hourly Settlement migration: ≈0 meters in November 2025, 2m by February 2026, 7m by April, 20.6m by August — about 70% of GB.

December always dips — look at December 2024, which recovers to the highest gap of the series by the following March. December 2025 dipped and never came back. The dashed line is the 23:30 fingerprint arriving.

The fingerprint is at 23:30. Octopus's month-mean volume neither rose nor fell at 23:30 through all of 2025. Then it climbs monotonically through 2026 — +0.01, +0.04, +0.07, +0.08, +0.14, +0.14, +0.19 — tracking the migration curve. Every other supplier's volume falls at 23:30, as you would expect late at night. Octopus's rises, because that is when its EV customers plug in and settlement can finally see it.

Octopus front-ran the reform. Elective half-hourly settlement of its own domestic smart meters went 5,900 meters in 2021 Q1 → 190,000 in 2023 Q1 → 624,000 by 2025 Q3. It was paying to be measured years before it had to be.

Act 4 — the peak that wasn't there

Thirty-seven consecutive months, working days, one settlement run throughout. Octopus crosses down through the entire household pack. Every other majority-domestic supplier moves less than 0.6 points in three years, and none of them accelerates at migration.

Octopus has gone from the peakiest household book in GB to the flattest. Its evening 16:00–21:00 share fell from 28.58% in July 2023 to 25.57% in July 2026. On matched twelve-month windows the move is 2.20 points. Every other majority-domestic supplier moved less than 0.6 points.

The rate is the tell: −0.75 points a year before MHHS, −1.70 a year during migration — 3.5 standard deviations from Octopus's own pre-reform baseline. Nothing else in the residential group passes 2.

Octopus's customers were already flatter than the standard profile. Profiling could not see that, so it settled Octopus against a national evening peak its customers did not have. Octopus's customers were cross-subsidising peakier books. Migration is what stops it.

The price curve, in its own panel, is why shape costs money: +£17.30/MWh in the evening peak, −£11.00/MWh overnight, working days 2024–26.

And here is what it is worth. Britain's wholesale evening peak runs £17.30/MWh above the day's average and the overnight trough £11.00/MWh below it — a spread of £28.29. Octopus's implied 2026 volume is 30.5 TWh, so one percentage point of evening share is 305 GWh a year. The 2.20-point move is £11.6m a year at the evening premium, £19.0m against the full spread.

Three separate mechanisms are in that chart, and only one of them is load shifting:

• 06:00–16:00 — change: −1.7 pp · what it is: generation, not shifting — solar netted into the same BM unit

• 16:00–21:00 — change: −2.2 pp · what it is: the peak shift — batteries and smart tariffs, plus MHHS revealing it

• 23:30–06:00 — change: +4.0 pp · what it is: EV and battery charging

Only the evening move is worth the £11.6m. Octopus's export MPANs went from 88,000 to 293,000 over the same window, which accounts for the midday fall on its own. Getting this wrong — crediting the whole 4-point move to load shifting — would roughly double the number.

Two of those three mechanisms are still running and one has finished. The solar gap is now enormous and static: Octopus took 16.5% of its summer day between 10:00 and 15:30 last year and 16.6% this year, against 21.3% and 21.2% for the rest of the household pack. Panels made Octopus's summer look different, and they did that some years ago. The overnight move has not stopped — and it shows up most starkly in the season where Octopus still looked ordinary. Between 23:00 and 05:30 last winter Octopus took 17.2% of its day; this winter, 19.6%. The other eight went from 16.3% to 16.0%. A book that was within a point of everyone else overnight in the dead of winter is now two and a half points clear of it, and that is the half of the day where there is no sun to explain it.

Honest limit: this values the shape at wholesale prices. It is what the volume costs to buy, not a transfer anyone writes a cheque for. And splitting the move into "behaviour" and "measurement" assumes the pre-MHHS trend would have continued, which is an assumption rather than an observation.

Act 5 — everyone else did something different

Same two twelve-month windows, same price shading. The other suppliers are not smaller Octopuses.

• Octopus — night: +4.0 · midday: −1.7 · evening: −2.2 · peak shift, EVs and solar at once

• British Gas — night: +1.5 · midday: −0.9 · evening: −0.8 · the same three moves, a third as far

• E.ON — night: +1.8 · midday: −2.1 · evening: −0.0 · solar and some overnight, evening unmoved

• OVO — night: +0.8 · midday: −1.6 · evening: +0.2 · solar, but no peak shift at all

• ScottishPower — night: −0.4 · midday: −1.8 · evening: +1.6 · peak went up

• Utilita — night: +0.3 · midday: +0.3 · evening: −0.6 · almost nothing moves

Every book except Utilita's loses midday volume. That is GB-wide embedded solar, not any supplier's strategy — E.ON's −2.1 points is the largest of the six.

ScottishPower is the control. It is the only large household book whose peak went up, and it is not a settlement effect at all: its non-domestic volume fell from 2.45 TWh in 2022 Q1 to 0.77 TWh in 2026 Q1, down 69%, taking its domestic share of volume from 55% to 73%. Strip out the flat 24/7 industrial load that used to dilute the peak and the household peak stands out more. Nothing about its customers changed; the denominator did. That is what a shape change looks like when the cause is ordinary commercial retreat — gradual, starting years earlier, no acceleration at migration. It is the contrast that makes the Octopus result convincing rather than merely striking.

Act 6 — is Octopus actually greener?

2026 averages. Grid carbon intensity from NESO; rooftop solar from each supplier's domestic export MPANs at that supplier's own mean install size, taken from Ofgem's SEG installation dataset, scaled by national PV output.

So far this has all been about when people use electricity. The obvious next question is whether any of it is cleaner — and the obvious answer is wrong twice over.

On grid draw alone — which is all settlement sees — Octopus looks the worst of the lot. Its customers pull 134.2 gCO₂/kWh against British Gas's 132.8 and EDF's 132.5. Every domestic supplier is above the grid average of 130, because households are peakier than the system and peaks are gas. But Octopus is worse than most, for a perverse reason: its customers self-supply through the middle of the day, so they skip the grid's cleanest hours. Per unit taken from the grid, what is left is dirtier.

A second effect pulls the same way, and it is the subject of Act 7: Octopus's book has more electrified heat than any other major. Heating runs on winter evenings — the dirtiest hours of the dirtiest months. So the two things this supplier is doing that you would most want a supplier to do, putting panels on roofs and heat pumps in houses, both push it up the only carbon measure settlement can see.

Add their own generation back — settled volume is consumption minus generation, so consumption is settled plus generation — and the ranking inverts. Octopus goes to 127.9 gCO₂/kWh, the cleanest book in Britain, and the only one whose customers' electricity is, on an annual average, cleaner than the grid they draw it from.

On an annual average is doing a lot of work in that sentence. Solar is seasonal, and violently so: Octopus's rooftop fleet removes 11.1 gCO₂/kWh in July and 1.1 in December, a tenfold swing. Take it month by month rather than as a rolling year and Octopus sits below the grid from March to September and back above it from October to February. The advantage disappears in precisely the months it would matter most — when the grid is dirtiest, demand is highest, and, for a book with more electric heating than anyone's, its customers are using the most.

But it is a lead built on stock, not speed. Measured on how much each book has closed on the grid since September 2023, British Gas has improved faster than Octopus — −2.5 gCO₂/kWh against −2.1, and on grid draw alone Octopus is fifth of nine and barely moves at all. Its own solar makes its customers skip the grid's cleanest hours, which cancels most of what it gains by shifting load off the evening peak. Octopus is far ahead because it acquired the panels, not because it is pulling away.

The gap between the two measures is the cleanest single statistic in the piece: gCO2/kWh removed by a supplier's own rooftop solar, twelve months to July 2026.

• Octopus — gCO₂/kWh removed: 6.30 · behind Octopus: —

• ScottishPower — gCO₂/kWh removed: 1.88 · behind Octopus: 3.4×

• British Gas — gCO₂/kWh removed: 1.06 · behind Octopus: 5.9×

• E.ON — gCO₂/kWh removed: 0.99 · behind Octopus: 6.4×

• So Energy — gCO₂/kWh removed: 0.96 · behind Octopus: 6.6×

• OVO — gCO₂/kWh removed: 0.87 · behind Octopus: 7.2×

• Utility Warehouse — gCO₂/kWh removed: 0.79 · behind Octopus: 7.9×

• EDF — gCO₂/kWh removed: 0.23 · behind Octopus: 27×

• Utilita — gCO₂/kWh removed: 0.03 · behind Octopus: 201×

So: ScottishPower is Octopus's closest competitor on greening its customers' electricity — at about a third. Nobody is close.

And now the deflation, because it matters more than any of the above. Six grammes per kilowatt-hour sounds like a lot until you put it against the number underneath it. Octopus's entire rooftop fleet — 314,000 installations, the largest in Britain by a factor of twelve — removes 4.7% of its customers' electricity emissions. British Gas's removes 0.8%. Utilita's removes 0.02%.

Over the same three years the grid itself went from 167 to 130 gCO₂/kWh: a 23% cut. The grid did six times more for Octopus's customers than Octopus's entire solar fleet did.

That is the proportion to hold on to. Everything in this article — the solar, the batteries, the smart tariffs, the peak shifted off the evening — is happening in the margins of a number being driven overwhelmingly by what is connected to the transmission system. It matters for who pays what, and it is a real and growing commercial divide between suppliers. It is not, on this evidence, how Britain decarbonises its electricity. That was done by building wind farms.

Install sizes come from Ofgem's Smart Export Guarantee dataset, which records every SEG installation's capacity by licensee. It is worth reading on its own: Octopus holds 314,470 of Britain's 405,439 SEG solar installations — 78% of them — and its share sits between 74% and 80% in every single region. Its systems are also the smallest of the majors, averaging 5.36 kW against British Gas's 7.70 and ScottishPower's 7.78: mass-market suburban roofs rather than large rural arrays.

One caveat that runs the other way, and it is a big one. This measures the carbon of electricity, and electricity is only part of a household's energy. A home that swaps a gas boiler for a heat pump uses far more electricity — which makes its supplier look worse on this chart — while cutting its total emissions sharply. Gas heat costs about 215 gCO₂ per kWh delivered once boiler losses are counted; the same heat from a heat pump at a seasonal efficiency of three, on a 130 gCO₂/kWh grid, costs about 43. Roughly a fifth.

That matters here because the suppliers with the most electric heating are the same ones that top this chart's grid-draw column. Octopus's demand rises 6.1% per heating degree day and OVO's 6.2%, against British Gas's 4.2% and EDF's 3.5% — they are carrying more of Britain's heat on the wire rather than in a pipe. An electricity-only carbon metric penalises exactly the thing you want suppliers to be doing. Read this chart as "the carbon of the electricity they bought", never as "the carbon of their customers".

Three further things it deliberately does not do.

It ignores REGOs and green tariffs, which are a contractual matter and tell you nothing about when anyone consumed.

It undercounts solar, everywhere. Panels installed under the Feed-in Tariff have deemed export — a flat assumption, never metered — so those homes hold no export MPAN and appear in neither Elexon's counts nor Ofgem's SEG dataset. Their generation still suppresses their supplier's settled volume; this measure simply cannot see it. GB domestic rooftop is usually put at 4.5–5 GW, and the metered export fleet modelled here is about 1.9 GW. Every solar figure in this article is a floor, and most of all for whoever holds the legacy FiT book — which, given when those panels went up, means the incumbents rather than Octopus.

And it does not tell you whether Octopus made its customers greener or simply attracted the ones who already were. A large part of that export MPAN curve is FiT-era homes re-registering onto metered export, not new panels going up.

Act 7 — where the gas grid runs out

Daily volume regressed on heating degree days with a time trend, by supplier and by GSP group. Only the majority-domestic books compare fairly with each other.

That last caveat deserves its own chart, because the thing the carbon metric penalises is measurable. Regressing each book's daily demand on heating degree days gives its exposure to electric heating — and it turns out to be the sharpest regional result in the piece. Business books come in at 0.3–1.1% per degree-day; household books at 4–6%. OVO and Octopus are the most cold-sensitive majors, at 6.2% and 6.1%, against British Gas at 4.2% and E.ON at 3.7%.

And this is one place where Octopus is not pulling away. Fit the same regression separately for each of the three years and every household book except two is getting more cold-sensitive, not just Octopus: it goes from 4.9% to 6.5% per degree-day, but Utility Warehouse goes from 4.9% to 6.9% and ScottishPower from 4.2% to 5.7%. Electric heating is arriving across the residential market at roughly the same rate. Whatever is distinctive about Octopus's book, its rate of electrifying heat is not it.

The obvious objection is that this is the solar talking: panels suppress a supplier's summer volume, which would make its winter look heavier without a single extra radiator. It is not. Restrict the same test to 22:00–04:00 — hours in which no British solar panel has ever generated, in any month — and the picture holds: Utility Warehouse's cold sensitivity rises 1.6 points over three years, Octopus's 1.1, and every other household book is flat or falling. Add the estimated solar generation back to the whole day instead, and Octopus's winter volume still grows relative to its summer, with panels accounting for well under a quarter of the move. Two suppliers are heating more homes with electricity. Neither of them is doing it because of the sun.

But the real pattern is geography, not company. Averaged across the household books, North Scotland runs at 8.3% per degree-day and South Wales at 3.5% — more than twice the sensitivity, in a country you can drive across in a day. Octopus's North Scotland book alone runs at 16.8%, the highest figure in this entire analysis.

And it is emphatically not simply "colder places use more heat". Southern England is more cold-sensitive than Northern England — 5.5% against 4.6% — while being over a degree-day warmer. South Wales, the least sensitive region in Britain, sits below London. What the map is actually drawing is the gas grid: where mains gas thins out, in the Highlands and in rural Devon, Cornwall and Wessex, heat runs on the wire. Where the pipes are dense — the Valleys, London, the northern cities — it does not. This is a map of pipes, not of weather.

Act 8 — a transition for homeowners

Run back through what this article has measured. Rooftop solar. Home batteries. Electric vehicles charged overnight. Heat pumps. Every single one of them requires two things: a roof you own, and capital to put on it.

Which is why the most instructive book in the whole dataset is the one that barely moves.

Utilita supplies about 710,000 households and is Britain's largest prepayment specialist — a segment much of the market has spent years trying not to win. Its book is 96% domestic by volume, so it compares directly with Utility Warehouse and So Energy, and on every measure in this piece it sits at one end:

• domestic export MPANs per 1,000 meters — Utilita: 0.23 · Octopus: 39.4

• SEG solar installations, all of GB — Utilita: 89 · Octopus: 314,470

• evening peak share, change over three years — Utilita: −0.6 pp · Octopus: −3.0 pp

• demand response per heating degree day — Utilita: 4.1% · Octopus: 6.1%

• electricity per meter per quarter — Utilita: 906 kWh · Octopus: 1,154 kWh

None of that is a failing of Utilita's. You cannot fit solar to a roof you rent. You cannot buy a battery, or an electric car, or a heat pump, on a prepayment meter and a tight month. The Smart Export Guarantee pays people who own a roof. Heat pump and EV grants reduce the cost of a purchase for people in a position to make one. Time-of-use tariffs reward households that can move consumption — which usually means households with an appliance expensive enough to be worth moving.

Read that column again and it is not a story about a company. It is the shape of a policy settlement in which the instruments of the flexibility transition are, almost without exception, available to people who own property and have money to spend on it. Utilita's book is what the transition looks like from the other side of that line.

One number deserves care rather than a punchline. Utilita's customers respond least to cold — 4.1% per heating degree day against Utility Warehouse's 5.8% on a book just as domestic. There are two readings and this data cannot separate them: those homes may simply have less electric heating, or they may be heating less when it gets cold. Self-rationing and self-disconnection among prepayment households in cold weather are well documented across the sector and are not specific to any supplier. If it is the second reading, a book that does not respond to a cold snap is not a book doing well, and that is a matter for policy rather than for the company that serves those customers.

What half-hourly settlement changes is that all of this stops being an inference. Until now the settlement record assigned these households the same national curve as everyone else — the same eight curves this article opened with. From now on it will show, in public data, month by month, exactly how far the flexibility transition has and has not reached the people it was hardest to reach.

That seems to me the most useful thing likely to come out of the reform, and it is not the eight figures a year.

Conclusions — the pack holds, one book leaves

Four things follow, in roughly descending order of confidence.

One: the market is not fragmenting. One supplier is leaving it. I went looking for eight books drifting apart and did not find them. Octopus's settled shape sat in the middle of the household pack three years ago and is now outside it on most axes measured here — the flattest evening peak, the most overnight load, ten times the solar density. Not all of them: it is second to OVO on cold sensitivity, and its rate of electrifying heat is unremarkable. But the other eight have barely moved relative to each other. Every one of them shifted less than 0.6 points of evening share in three years, against Octopus's 2.20, and none of them accelerated at migration. Several of them are closer to each other now than they were before migration began. This is not a market of near-identical retailers coming apart at the seams. It is a market of near-identical retailers, minus one — and the one is pulling away in the hours and the season where being ordinary costs the most.

Two: the averaging was an insurance policy, and it has been cancelled. Not by design, but in effect. When every household is settled against the same eight curves, each supplier is charged for the average customer rather than its own. A supplier whose customers were unusually peaky was carried by one whose customers were not, and neither could see it happening. Octopus spent years paying for an evening peak its customers did not have; its customers were subsidising peakier books. That is what profiling was — a mutual insurance scheme nobody signed up to, priced at zero, settling claims silently every half hour for twenty-five years.

Each supplier is now charged for the shape it actually has. Whoever has the flattest book stops paying for someone else's peak: worth eight figures a year to one supplier already, on one measure, one year into a transition that is 70% done. The reverse is the part that will hurt. Whoever has the peakiest book now owns that cost outright — and owning a cost is not the same as being able to do anything about it. You can no longer ride the pack. There is no pack.

That is the asymmetry, and it is where this gets interesting. Settlement now prices a supplier's shape by the half hour, while most retail tariffs still hide time from the customer completely: one unit rate, every hour of the day, all year. On Ofgem's most recent figures a majority of accounts at several large suppliers sit on price-capped default tariffs — 74% at OVO, 60% at ScottishPower, 57% at EDF — which are flat by construction. (Octopus and E.ON Next are absent from that release, which is exactly the wrong hole; see below.) Those suppliers have been handed a bill denominated in half hours and a customer relationship denominated in a single number. Nothing the customer sees tells them when to run the washing machine, so nothing the customer does will change what the supplier pays. The bill has arrived years before the lever.

Which makes this a selection event, and the trait being selected for is not the one the industry is organised around. Not scale, not brand, not the quality of your hedging desk. It is the ability to persuade several million people to move electricity into a different hour of the day — and increasingly to do it for them automatically, while they are asleep. That is a software and behavioural competence wearing an energy company's clothes. Octopus's 2.20-point move did not happen because it forecast prices better than British Gas. It happened because it built tariffs that pay people to shift, an app that makes the shift legible, and direct control of the car and the battery so that most of its customers never have to think about it at all.

The uncomfortable part for everyone else is that this cannot be bought quickly. You can acquire a book of customers, and several suppliers here have, at speed and at scale. You cannot acquire their willingness to be moved. That has to be built, tariff by tariff and app release by app release, over years — and the suppliers furthest behind are precisely the ones whose customers are most likely to be sitting on a flat default tariff they have never had a reason to leave. Three years ago that was a comfortable place to be, because the profile was doing the work. It is now a structural cost with a compounding interest rate.

Three: the cheapest way to flatten your book is to choose who is in it. Persuading millions of people to move their demand is slow, expensive and uncertain. Selecting customers who are already flat is fast, and it is free. And half-hourly settlement is what makes it possible: for the first time a supplier can put a settled, half-hourly cost against an individual customer and see what that customer is actually worth. Under profiling, a household's shape was invisible to the profit and loss account, because everyone was charged for the average one. It is not invisible any more.

Nothing crude has to happen for this to bite. No supplier needs to refuse anyone. You simply advertise the electric-vehicle tariff where electric-vehicle owners are, bid harder on the comparison sites that surface to homeowners, price the smart tariff attractively and the prepayment product adequately, and compete a little less hard to retain the customers whose evenings are expensive. Every one of those decisions is ordinary commercial behaviour, and in aggregate they sort the market by shape.

Then the same logic runs from the other direction, which is where it turns into a genuine unravelling. Flexible households will move to time-of-use tariffs, because that is where their flexibility finally gets paid for. Inflexible ones will stay on the flat default, where they are charged the same rate at six in the evening as at three in the morning. As the flexible leave, the flat pool gets peakier, and the average cost of serving whoever remains in it goes up. That is textbook adverse selection, and the exit is one-way.

Under a price cap, it does not surface as a rising price. The cap is set against a benchmark, so a supplier holding a book that is quietly getting more expensive to serve cannot simply charge for it. It surfaces as a margin, and then as a loss. Britain has already watched what happens to suppliers holding books they are not allowed to price, and it was not an orderly process.

And the residual pool is not a random draw. Every mechanism in this article — panels, a battery, a car charged at half past eleven, a heat pump — needs a roof you own and capital to put on it. The households that cannot move are renters, flat-dwellers, prepayment customers, people whose consumption is already low because it is rationed rather than optimised. Utilita's book, the one that has barely changed shape in three years, is the clearest example in the data, and it is not a failure of that supplier so much as a description of who its customers are. As measurement improves, being flexible gets cheaper to serve and being inflexible gets more expensive — and the people on the wrong side of that line are the ones already least able to cross it. Nothing in the reform causes this. It stops hiding it, and hiding it was the only thing keeping the costs pooled.

I should be clear that this is a reading of the mechanism, not an observation. Two years of settlement data show suppliers' books diverging; they do not show anyone deliberately selecting customers by shape, and I have no way to test that from P114. But the incentive is new, it is large, and it points one way.

Four: and that is still the best argument for having done it. You cannot fix a distribution you cannot see. For twenty-five years the question "which households are actually flexible, and which are not?" had no answer in public data, because the data was averaged before anyone could ask. From November 2025 it has an answer, published monthly, for free, for every supplier in Britain. Whether that gets used to design a fairer transition or simply to price the existing one more accurately is a policy choice, not a technical one — but it is now a choice somebody has to make, rather than a fact nobody could establish.

What I could not establish

Whether this is already showing up in what people pay. Ofgem publishes exactly the right series — domestic accounts by supplier, split into price-capped default tariffs and fixed — and the current snapshot has a hole in precisely the wrong place: "data as of April 2026 does not include Octopus Energy and E.On Next due to a reporting issue." The spread among those that did report is quoted in the conclusions above, and it is wide enough to be worth someone's time. But the supplier this article is largely about is missing from it, so the argument that flat default tariffs leave a supplier without a lever is a reading of the mechanism, not a measurement of it. Whether physical divergence has become commercial divergence is the obvious follow-up, and it needs one Ofgem spreadsheet that currently does not exist. There is no time series either — each chart in that release embeds a single snapshot, so the trend cannot be traced without going quarter by quarter through archived publications.

What this rests on, and what it doesn't

Five things a careful reader should know, all of which are in the piece already but are worth having in one place.

Volume, not customers. Domestic and business shares throughout are shares of energy. EDF is 24% domestic by volume and 80% by meters. Load shape is volume-weighted, so volume is the right denominator — but the phrasing means its energy, not its customer base.

Solar carbon savings are seasonal. The annual figures blend a tenfold swing: Octopus's fleet removes 11.1 gCO₂/kWh in July and 1.1 in December, so it is below the grid March–September and above it October–February. Any annual average here conceals that.

The carbon numbers are small. Octopus's solar removes 4.7% of its customers' electricity emissions; everyone else's removes under 1.5%. The grid decarbonised 23% over the same period. Supplier-level differences are real but marginal against the generation mix.

The carbon chart is electricity only. A household swapping gas for a heat pump buys more electricity, pushes its supplier up that chart, and cuts its total emissions by roughly four fifths. Act 7 is the counterweight to Act 6 and they should be read together. No adjustment for displaced gas is attempted: the data cannot separate heat pumps from resistive heating, whose credits differ threefold, and the largest electric heating loads are off the gas grid, where nothing is displaced at all.

Solar capacity is measured, its output is modelled. Install counts and sizes come from Ofgem's SEG dataset and Elexon's export MPAN counts; the half-hourly shape is national PV output scaled to each fleet. FiT-era homes, whose export was deemed rather than metered, appear in neither count — so every solar figure here is an undercount, and most so for whoever holds the legacy FiT book.

Behaviour and measurement are entangled. Splitting Octopus's 2.2-point peak move into "customers changed" and "settlement started seeing it" rests on assuming the pre-reform trend would have continued. It is an assumption.

The £11.6m is a purchasing cost, not a transfer. It values the shape at wholesale prices. Nobody writes that cheque; it is what the volume costs to buy. The per-party settlement cashflow records were parsed and do not contain it — under profiling a supplier hedges to its profiled volume, so a wrong profile generates no imbalance.

Sourcing

All public: Elexon P114 S0142 files via a portal scripting key, the Elexon MHHS dashboard, Elexon's Gross Supplier Market Share reports, the BMRS BM Unit registry, BMRS market index prices, the NESO Carbon Intensity API, Sheffield Solar PV_Live, Ofgem's Smart Export Guarantee installation dataset, and ERA5 temperatures via Open-Meteo. No commercial data.

In 2025 Great Britain imported 21.9 terawatt-hours of electricity from France. That is 9.3% of everything the transmission system supplied, arriving down three cables with a combined capacity of 4 GW. So far in 2026 the share is 10.3%, and power has flowed from France to Britain in 96% of all half-hours.

Some ways to feel the size of that. It is 61% of what the entire British nuclear fleet generated last year. It is more than every solar panel in the country produced — 19.8 TWh of GB solar against 21.9 TWh of French imports. And it is 84% of the output Hinkley Point C is designed to deliver — a station costing something like £35bn in 2015 prices to build, and not expected before 2030.

This is not a trading relationship any more. It is a supply.

Read more

I know exactly why that number lodged in my head. It is because my electricity bill is almost nothing. We have solar panels, a battery, a heat pump and two electric cars, and between them they have taken a bill that would otherwise be one of the largest things we pay and reduced it to a rounding error. Every one of those was…

Read more

Britain's solar fleet responded with a record of its own: 15.9 TWh generated by 16 August, more than any year before it.

Then the odd part.

Measured per panel, 2026 was unremarkable. Each installed megawatt returned 676 MWh over the year to date — only the sixth best in fourteen years, and barely above the long-run average of about 665. The record came from having more panels, not from better weather. Britain's solar growth is a build story, not a sunshine story.

Push on the summer alone and it gets stranger still.

Read more

You know what the holidays have done to your mornings. Here is what they have done to the grid's.

Read more

That is a measurement problem, and measurement problems can be fixed by measuring things.

The question I get asked most about this — and lately it has been asked with a certain amount of heat — is some version of: how hard is it, really, to forecast the grid? Surely we can tell how much wind and solar will be running tomorrow?

Mostly we can, and the people who do it are good at it. But the question carries an assumption, and it is the one thing here that has stopped being true. Forecasting assumes you find out what happened. Every forecast is trained, tuned and scored against outturn: you learn what you got wrong, and that is how the next one gets better.

For the metered half of the fleet, that loop works. A transmission-connected wind farm's output arrives five days after the fact, per unit, from a meter. But for the categories growing fastest — the small sites, the rooftops, the panels behind a meter with a load sitting next to them — there is no outturn at all. Not late, not coarse. Absent. What gets published as yesterday's embedded solar is a model's estimate of yesterday, upscaled from a reference sample of systems that are metered.

It is a good estimate, built by people who know exactly what they are doing, and I said so in part one. But you can only score a forecast against it by comparing one estimate with another, and both lean on the same weather data. The error you measure that way is not the error you have.

So the honest answer to "how hard is it to forecast?" is that the forecast is the tractable part. The hard part is that for a growing slice of the system nobody ever establishes what happened, and it is difficult to systematically improve a prediction of a quantity that is itself inferred. You cannot converge on a number that nothing ever pins down.

Read more

A new government is under pressure to answer that, fast, and without rerunning 2022's blank cheque. The answer currently on No.10's desk is a block tariff — a cheap first slice of energy for everyone, dearer units above it — pushed by the New Economics Foundation and the Joseph Rowntree Foundation, and with friends in high places: the Energy Secretary used to run NEF. Against it, the Centre for Net Zero published a piece last week titled "The wrong kind of radical," arguing block tariffs break price signals, punish electrification and mistarget the poor. (The Resolution Foundation, usually filed on the "pro" side, actually wants neither — their proposal is an income-targeted social tariff, a different animal.)

I think both camps are right about each other and wrong about the question.

Read more

It isn't. There is no such meter, there never was, and the further you look into how the number is actually assembled the less it resembles a measurement at all. Some of it is metered. Some of it is inferred. Some of it is a parameter somebody set once. And a growing share of the electricity Britain generates is now invisible to it by construction — not because nobody has got round to metering it, but because the way the system is defined means the quantity is never formed in the first place.

This is not a scandal. Every one of these choices was reasonable when it was made. It is more interesting than a scandal: it is what happens when a measurement framework built for about a hundred large machines is asked to describe a system with two million.

It is also, I think, the reason for a question I get asked more than any other, usually in some version of: how good are we actually getting at forecasting wind and solar? There is a fair amount of heat in that question at the moment.

My honest answer is that the forecasting is better than most people assume and the measurement is worse — and that the two are joined at the hip, because a forecast is only ever as good as the outturn you score it against. Here is the tell, which gets a section of its own further down. Britain's day-ahead wind forecast runs about 14% high, on five days out of six. That sounds like a forecasting failure. It is not one. The forecast is answering a question nobody meant to ask it, and working out which question takes you directly into how demand is defined — which is where the rest of this piece goes.

So this is not really an article about forecasting. It is an article about what a forecast is being compared against, and how much of that comparison is estimated rather than measured.

The old world, where this all made sense

Start where the definitions started. In the 1990s, almost everything that generated electricity in Britain was a large machine bolted to the transmission system: coal stations, a handful of nuclear plants, the first big CCGTs. A few dozen sites, all of them individually metered, all of that metering read centrally and in real time by the system operator.

Read more

Solar is a victim of its own success

More and more solar power depresses (cannibalises) the price it gets from straight selling the power it generates to the grid. I like to decompose this between the "within day" effect, and the broader seasonal effect of summer (being sunnier). British solar, including all the older RO projects (potentially on borrowed time on a PPA) have been spared the worst seen already on continental Europe, but since 2022 the "captured price" (the average weighted price solar can get) has started to fall consistently.

Read more

The figures being quoted for it are wrong, though, and not by a little. Getting to the real ones took three corrections, and every one of them is a better story than the record itself.

The first is that a large slice of British wind, and very nearly all of British solar, is invisible to the dataset most people quote. It is real electricity, metered somewhere else, and simply absent from the series that gets cited.

The second is that a growing share of the wind was generated and then paid to stop. That volume can be reconstructed settlement period by settlement period, and putting it back changes how fast the fleet has actually grown.

The third is the uncomfortable one. The Elexon dataset built for precisely this question is wrong in both directions, for months at a time — and the error changes sign partway through, right across the years you would use for a year-on-year comparison. Nothing flags it. Every figure is present; they are just wrong.

Below, working from eleven years of half-hourly generation data, every balancing action taken on a wind farm since 2016, and 1.14 million forecast-hours:

• what GB wind and solar really produced this year, counted properly

• whether the two genuinely hedge each other, and the trap that makes the answer come out backwards

• how good NESO's wind forecast is, once you stop grading it against the wrong thing

• the one weather regime where it reliably over-predicts — and why the obvious physical explanation for that turns out to be wrong

• what it would take to fix a British winter by building more solar, an arithmetic that ends the argument

• and how long Britain's wind-and-solar droughts actually last, which is the only number that sizes storage

Read more

The short version: CfDs are better than their critics say and worse than their defenders say, and the thing that trips up almost everyone — me included — is generalising from one contract to all of them.

Here is the number that makes the point. Between 1 January and 15 July 2026, the levy paid £1,365.8m to 81 generating units across 20.2 TWh. That is an average top-up of £67.63 per megawatt-hour.

Two of those 81 units are within £10 of that average.

Read more

Read more

Each panel shows weekly reservoir filling as a share of that market's peak, with 2026 in blue against the 2016–25 range. The pink line is 2022 — the last time Europe had a hydro year bad enough to make front pages.

In 2022 the drought was southern. Spain and Portugal spent that whole year pinned to the bottom of their ten-year envelope — Portugal flat around 30% through the summer — and Italy's Po valley crisis put its line on the floor of the band too. The Nordics, meanwhile, had a roughly normal year.

In 2026 the picture has flipped. Norway and Sweden drew down through the spring along the very bottom of their decade range — Sweden touched its ten-year minimum in April, and both dipped deeper than they did in 2022. The snowmelt refill has since pulled them back towards the median, but the hydrological hole peaked at more than 25 TWh across the Nordic region. At the other end of the map, Spain and Portugal are hugging the top of their ranges, and Greece has been above its interquartile band all year.

Read more

Read more

In May I argued that plug-in solar is a red herring. That piece has found a second audience over the past few weeks, and the questions in my inbox have converged on one: should I buy a kit when the law changes?

The law changes on Thursday 27 August. The bank holiday falls on Monday the 31st. Four days — and you already know how it will look: a boxed kit in the supermarket middle aisle, wedged between the pressure washers and the padded ski jackets, at a price pitched to feel like now-or-never.

So here is the direct answer, and it is not the one the retailers want: don't. Partly because, as you'll see, there is nothing yet to buy. But mostly because the thing they'll eventually sell you is the wrong half of the story — and the right half is sitting in a regulatory waiting room.

The shelf is empty

Start with the fact that makes "rushing" impossible. A compliant plug-in kit has to appear on the ENA's public type-test register before it can be sold, and the register has a filter for exactly this product. Set the Type field to "Plug-in Solar" and here is what you get today, five weeks out:

Zero devices of zero. The shelf is built, labelled and empty.

It has been coming for months, in the loosest sense of the word. The government announced its intention to legalise plug-in solar on 16 March, in a post-Iran energy-security package. The instrument that did it, SI 2026/848, was made on 16 July and comes into force on 27 August — 122 days from announcement to signature, for a document of 1,103 words. That's nine words a day; the law is shorter than this article. It was also finished in a hurry at the very end: DESNZ published its consultation response six days ahead of schedule and made the regulations the same day, 16 July — four days before Keir Starmer handed the premiership to Andy Burnham. Whether by chance or design, plug-in solar was got over the line just as the government changed leader.

But the drafting pace isn't really the story, and it would be lazy to pretend it is. The March announcement wasn't a launch, it was a statement of intent — and it was made for a season the scheme was never going to be ready for. Nobody who heard it was ever generating this summer, because five months on there is still nothing certified to buy. You trail a product at the start of the season that sells it, then spend that whole season not shipping it. The first plug-in summer that actually exists is 2027.

Even once it's stocked, the panel is the wrong buy

Say you wait for the register to fill. What are you buying? Less than the marketing implies, and here the May argument only hardens.

It earns almost nothing, because it can't sell anything back. A plug-in kit can't be MCS-certified, and without MCS there is no Smart Export Guarantee. So every kilowatt-hour you generate and don't use in the same moment is a gift to your supplier. Anchored to my four years of measured wall-mounted yield and modelled against real half-hourly household demand, a typical home self-consumes about 64% of what a small kit makes and spills the rest. Value it at the price cap and an 800 VA kit is worth roughly £75 a year — against the £130–150 I casually put in the May piece by valuing every kilowatt-hour at retail. The no-export rule lops about a third off the top, and it bites hardest in the bright spring and autumn months, when the panel makes most and the house soaks up least:

To put that in perspective: a whole winter's generation from one of these — September to February — is worth about £24. So even once there is finally something to buy, the month you choose to buy it barely moves the needle. The thing itself is cheap; the timing of it is a rounding error.

So what should you actually pay? Work it backwards, and treat this as the discretionary purchase it is: demand a 25% return a year — a four-year payback — which is about right once you allow for the cost of the money, the kit ageing and dating, and the fair chance that you (a renter especially) move before it clears. On that hurdle the most you should hand over is roughly four times what it saves you in a year. At the £75 a typical wall kit earns, that's £300; a perfectly-sited, always-occupied home earning nearer £120 might stretch to £480. Either way, the kits on sale today — £499 and climbing — are already above the ceiling a clear-eyed buyer would set. And that is a rule that ages better than any single figure of mine: whatever a kit ends up costing, ask whether it will save you a quarter of it, every year. If not, it's too dear.

One honest exception, and it's the nice one. That sum values your time at nothing, and installing a kit isn't nothing: budget a few hours to mount it, fit the plug, notify your network operator, and nudge the angle across a few weekends to find where it likes to sit. If that sounds like a good rainy-Sunday project — and it genuinely can be — call the labour a hobby and forget it. If it sounds like a chore, add it to the bill, and the case gets thinner still.

And if what moves you is carbon rather than cash, the comparison is crueller. On today's grid — about 120 grams of CO₂ a kilowatt-hour, and falling fast — that kit's 450-odd kilowatt-hours a year displace roughly 55 kg of CO₂, shrinking a little every year as the grid itself cleans up, and shrinking again once you count the carbon it took to build the panels. Call it well under a tonne across the kit's entire life. Now take the same £300 — or £500 — and spend it instead on UK or EU carbon allowances, and cancel them: at 2026 prices near £70 a tonne, you lift four to seven tonnes straight out of the emissions cap, today. Pound for pound, tearing up carbon permits abates several times more than bolting a panel to a wall, and it does it now rather than dribbled across twenty years. Be as generous to the panel as the physics will allow — credit it with displacing marginal gas rather than the average grid mix, which roughly triples its lifetime saving to a couple of tonnes — and the permits still come out ahead, and still sooner. And sooner counts for far more than it sounds. What warms the planet is the running total of everything ever emitted, so a tonne kept out of the air this year is worth more than the same tonne kept out twenty years from now: it is cut for longer, and it holds the peak lower while the tipping points are closest. The panel delivers its carbon the slowest way imaginable — a thin trickle over two decades — exactly when the atmosphere would most value it up front. That is a deflating thing to know about a product sold on its greenness. A cancelled permit doesn't glint on the balcony, granted, and it won't teach you a thing about your own demand — but if the ledger you care about is tonnes of carbon, the wall is not where the money goes furthest.

The cap tops it out. 800 VA is a ceiling on generation, and it clips every sunny day, so a plug-in panel makes a few hundred kilowatt-hours a year no matter how much silicon you bolt to the wall. The spec lets you hang 2,000 W of panel behind that 800 VA inverter — a 2.5-to-1 ratio where a rooftop system runs 1.1 or 1.2 — but with no export payment and no battery allowed, every watt above what you're using in the moment is simply thrown away. You are permitted to buy panel capacity that is structurally unable to earn.

And Britain made it dear at the bottom. Not at the top: EcoFlow's STREAM kit is about £499 here and near enough the same on its German store, so the branded mid-market is level. The difference is the floor. In Germany the cheapest option isn't a £499 branded set — it's a €249 own-brand kit from Lidl, or a self-assembled build for less again. That long cheap tail is what made balcony solar a mass-market object there. Britain's rules abolish it by construction: complete certified kits only, so no self-assembly; whole-kit certification and a register-before-sale gate, so the discounter's €249 box can't clear the bar the way a self-declared German product does. The premium isn't that any given kit costs more here — it's that the £150 and £250 options that dominate Germany can't legally exist, so the market starts at the branded price and stays there.

So ignore the German paybacks. You'll see two-and-three-year figures quoted at British readers; they describe a different planet. Germany's smart-meter rollout has all but failed — 5.5% of homes — so roughly half the country still runs an old spinning-disc Ferraris meter, and since 2024 the law lets that disc run backward until the operator swaps it. That's free net metering at the retail rate, read once a year, often by the householder. And the same missing smart meter that hands them free export also locks them out of dynamic tariffs, so for most Germans a panel is the only way to touch cheap solar at all. A British buyer has neither problem and neither perk: a smart meter that reports every half-hour, no export payment to collect — and, crucially, the ability to just switch to a tariff like Agile and buy the grid's cheap midday solar without bolting a thing to the wall.

And there is a safety case worth taking seriously, because it came from the one expert quarter that dissented. When the government consulted, 88% of the public waved plug-in solar through — but the bodies who actually understand home wiring did not join the cheer. The ECA, Electrical Safety First, the IET, NICEIC and SELECT issued a joint warning, and NAPIT another: not against plug-in solar, but against rushing it. Their central worry is backfeed — a generator pushing power the wrong way into a ring main, past protective devices that a tired 1970s installation was never fitted to handle.

Follow that worry to its end, though, and it points away from the panel and toward the battery. A solar panel is an uncontrolled source: it pushes out whatever the sun gives and exports whatever you can't use, in the moment, whether the house is ready or not. A plug-in battery is a controlled one — it can meter itself to your own load and needn't spill a watt back to the grid, sitting as a buffer between your house and the network rather than a tap plumbed straight into it. It answers the very concern the panel raises. The people warning loudest about plug-in solar are, without quite meaning to, making the case for plug-in storage.

That's the whole case against the panel, and it's the May case with the numbers filled in. But it's only half the piece, because the interesting product isn't the panel at all.

Read more