Methodology

What each figure on the live dashboard means, how it is derived from the public feeds, and where the data can mislead you.

The live figures cover Great Britain, not the whole UK: why, on the About page →

About this dashboard

A live read of the GB transmission system from Elexon’s Insights (BMRS) API — no key needed. Your browser fetches the figures from Elexon, NESO and Sheffield Solar through this site’s relay, or directly if the relay can’t be reached. Every fuel row except solar is BM-metered. Embedded (distribution-connected) generation is invisible to those feeds and surfaces instead as suppressed National Demand: INDO on a sunny afternoon looks implausibly low, and that is the most common misreading of this data. Embedded solar is filled in from PV_Live and counted; embedded wind is not. Biomass counts as low-carbon by UK convention despite being combustion. Pumped storage and interconnectors go negative when pumping or exporting, and are excluded from the low-carbon denominator.

Coal is not named as a fuel. Ratcliffe-on-Soar, GB's last coal station, closed on 30 September 2024 and COAL has read 0 every settlement period since. The code stays mapped so a residual could never be dropped silently, and the table still shows the raw zero — but naming it in a label would imply an option the grid no longer has.

“Other” is a residual, and cannot be broken down

Other is not a fuel. Elexon's own definition, in the BMRA and SAA Interface Specification, is circular by design: “all generation that cannot be placed within any of the other categories.” At 0.2–3 GW, peaking in the evening, it can exceed hydro, pumped storage, OCGT, oil and coal put together — not a rounding error.

It cannot be decomposed from published data, and this page checked. The feeds that carry these volumes carry no fuel type per BM unit, and the register that does carry one is self-declared at registration, by a different body over a different population — joining them would be inventing a mapping nobody publishes. The usual guesses are worth naming as guesses: waste-to-energy is largely distribution-connected and outside this feed entirely, and Other has never gone negative on a day when pumped storage did, which is not what net-charging batteries would look like.

Embedded solar, and why it is drawn differently

GB solar is almost entirely distribution-connected, so it is absent from every Elexon metered feed. Two sources publish it and they disagree, so which one you are looking at depends on how you opened this page — the solar row names its own source. Read that before quoting a solar figure.

Preferred: Sheffield Solar's PV_Live, the recognised measure of total GB PV outturn, about ten minutes behind real time. PV_Live sends no CORS headers, so the site fetches it through a small proxy of its own (the body is returned untouched; the source is credited on the About page). If that proxy cannot be reached, the page falls back to Elexon AGWS — the ENTSO-E B1440 flow, published about 2h30m late, which has read 16–29% below PV_Live at every half hour compared — and the solar row says so. Treat an AGWS figure as a floor, not a reading.

Embedded solar is counted in every total, because it is real generation powering the country, but always drawn as a hatched band, because it is modelled rather than metered. Provenance is carried by the fill, not by leaving real megawatts out of the arithmetic; where both sources are readable, the gap between them is shown rather than reconciled.

Underlying demand adds embedded solar back to INDO; embedded wind is not added. NESO publishes an Embedded Wind and Solar dataset, but what it publishes is a forecast rather than a measurement, and folding that into a demand figure would undo the measured-versus-modelled distinction the rest of this page rests on. Those gigawatts stay hidden inside demand: a choice about what this tool is, not an absence of data.

Daylight shading and the sunrise/sunset marks are computed, not fetched — NOAA's solar-position algorithm for a PV-weighted GB point (52.4°N, 1.5°W). Sunset across GB spans about 45 minutes end to end, so read them as representative rather than exact.

Carbon intensity, and a ruler that moves

The band names are re-cut every 1 January. As the grid decarbonises NESO lowers the thresholds, so the same reading slides down the scale: 95 gCO₂/kWh was low under 2025's cuts and is moderate under 2026's. The figure did not move — the ruler did. Each reading is banded on the row for the year it belongs to, and the panel says which year it used. The band is computed from the figure rather than read from the API's label, because settlement reruns revise the figure without always recomputing the label; where the two disagree, the panel says so rather than quietly choosing.

Half the time the headline number is a forecast. NESO’s actual for a half hour in progress does not exist yet, so any live carbon figure is a prediction until the half hour has finished — hence the actual or forecast chip on the face of the number.

System frequency, and two limits that are not the same

Frequency is Elexon's FREQ dataset at its native 15-second resolution — by a wide margin the fastest-moving figure here.

The panel draws two pairs of lines, deliberately differently: the statutory pair solid, the operational pair dotted. 49.5 and 50.5 Hz are statutory: the Electricity Safety, Quality and Continuity Regulations 2002 permit a variation not exceeding 1% either side of the declared frequency, and Grid Code CC.6.1.2 restates it. 49.8 and 50.2 Hz are operational — NESO's own standard, not a legal limit. Drifting outside the operational band is ordinary and happens most hours; leaving the statutory range is not.

Forecast convergence, and the split

The wind panel draws three series, all BM-metered wind on one MW axis: outturn, the latest forecast for each settlement period, and the first forecast ever published for it, typically two days ahead. The interesting question is not only how wrong a forecast was but when it stopped being wrong, and the gap between the two dotted lines is that answer. Both vintages are scored on the same periods, so the improvement is like-for-like. Some days the latest forecast is worse than the first, and the panel says so.

The onshore / offshore split comes from the same AGWS call as the fallback solar figure, and carries its own timestamp rather than the chart's, because AGWS lags.

Supply and demand never quite match

The band at the top of the dashboard opens on the gap between supply and demand, because the gap is the interesting part. The two sides are measured at different boundaries, by different datasets, on different cadences, so they disagree by a few hundred megawatts almost all of the time — a fact about measurement, not about the grid running short.

Which two figures are compared is the whole correctness of it. The balance compares electricity on the grid with underlying demand, not with National Demand, because both include the same estimate of embedded solar — so it cancels, leaving metered generation plus net imports less national demand, which is a real quantity. Comparing against National Demand would leave the embedded estimate standing in the answer: several gigawatts on a sunny afternoon and nothing at midnight, which is why the wrong pairing looks perfectly correct in an evening screenshot.

Expected output at each power station

The Power stations and Meet the fuel types pages show expected output, not metered output. Every balancing-mechanism unit tells the system operator, half hour by half hour, what it expects to generate — its physical notification (PN) — and the most it can export, its maximum export limit (MELS). The pages take the lower of the two for the current half hour, never less than zero; a battery can import as well as export, so its figure is kept between its import and export limits.

Metered output for each unit is settled days later, so these figures will not match the Generation page’s five-minute totals to the megawatt. A notification can run ahead of what a unit delivers; taking the lower of the two figures, rather than the notification alone, gets closer.

Reading a figure that is not there

Every headline figure is in one of four states, each drawn distinctly:

• A number — a real reading, at full weight.
• A dash — the figure has not arrived, or the feed did not answer on the last refresh. Hovering says which, and names the feed that failed.
• A number with a clock — a real reading, older than the refresh period it belongs to. The mark carries its age.
• Zero — a genuine zero: a fact about the system, not a gap in the data.

The distinction being protected is the last one: a missing value must never look like a real zero. A dash cannot be read as “none”, and a zero is never dressed as an absence.

Every figure names its source

Separately from whether a figure has arrived: what kind of figure it is. A figure with no mark is measured; an estimate is hatched; a forecast is dashed. Six states in all:

• Measured — metered outturn, and unmarked. Most of this site is measured; the absence of a mark is the statement.
• Estimated — hatched: modelled rather than metered. Embedded solar (Sheffield Solar’s PV_Live, with Elexon’s AGWS as the fallback). A total or a share with that estimate inside it says so beside the figure: incl. est. solar.
• Forecast — dashed: every forecast line on every chart is drawn with the same dash, and a figure that is a forecast says “forecast”. A limit or a comparison line is never dashed.
• Calculated — written as an equation: the terms listed above a rule and the result under an equals sign. Derived from other figures here rather than read from a feed. It carries no colour, because derived is not less trustworthy — and no chip either, because a chip beside an equals sign only repeats it.
• Stale — a clock and an age.
• Unavailable — a struck-through chip and the word. The only mark that reads as a negation, because it is the one state that must never be mistaken for a value.

Where a mark can say it, the prose does not have to.

Deliberately left out

Battery storage in the fuel mix. Elexon reports storage discharging but not charging, and showing discharge alone would count the same energy twice — once generated, again on the way out. It is left out of the mix rather than shown as a half-truth, and is not folded into “Other”. What the batteries are doing is on the Batteries page, unit by unit.

History on the dashboard. The dashboard shows the last 24 hours of most things and the last hour of frequency; the longer view is on the section pages — Generation keeps a daily record back to 2016.

Embedded wind, as above, including all three wind series and the onshore/offshore split, which are transmission-metered only.