How every number
is worked out

What is original here, and what is not

None of the underlying data is ours and none of it is secret. Elexon publishes every balancing action, DESNZ publishes the planning database, NESO publishes the constraint costs and the connection queue, ONS publishes the boundaries and the census. Anyone can download all of it. If you already work in energy, you have seen these sources.

What did not exist before, as far as we can tell, is the join:

• The constituency view. Elexon reports by balancing-mechanism unit, REPD by site with coordinates, NESO nationally. Nobody publishes any of it by Westminster seat. Putting money and megawatts into the geography that MPs, councillors and local papers actually work in required the hand-maintained BMU→site mapping (published, with confidence grades) and a point-in-polygon geocode of every site.
• The reason join. "Approved and not built" is in REPD; why it is not built is not. Joining the planning record to connection queues and appeal status, and reducing that to one plain sentence per project with a named decision-maker, is the part that turns a number into something a person can act on.
• The time series and the cleaning. A year of settlement-quality switch-off costs per unit, reconciled against NESO's own published volumes, with the sign conventions and the unit traps (their volumes column is MW per half-hour, not MWh) handled and tested.
• The translation. One comparator, used everywhere, for every pound; homes-equivalent stated as annual energy rather than instantaneous capacity; caveats at the point of use rather than in a footnote.

So: the facts are public, the arrangement is the work. If you think a join is wrong, the mapping file, the reconciliation and every seat's JSON are published so you can check it rather than take our word for it. Corrections are welcome and get credited.

Number 1 — £ paid to switch off wind farms (the industry word is "curtailment")

Source: the Elexon settlement bid stack (/balancing/settlement/stack/all/bid), which gives the accepted bid volume (MWh) and original bid price (£/MWh) per balancing-mechanism unit (BMU) per half-hour, as derived by Elexon settlement. We keep turn-down (bid) actions on wind BMUs only. Cost = volume × price: a wind farm bidding a negative price is paid to reduce output, so the product is a positive cost to the system.

Attribution: BMUs are matched to sites in the government's Renewable Energy Planning Database (REPD) by a hand-reviewed mapping file with a confidence grade per row (candidate suggestions are published too). One site can span several BMUs (Viking has four) and one BMU can span several REPD entries (phased builds — the share column splits them). Sites are placed in constituencies by point-in-polygon against the ONS July-2024 boundaries.

What is deliberately left out of seat totals: offshore wind (£210m in the window) is reported nationally, never attributed to a coastal seat. Money on BMUs we could not confidently map — £21m, 5.2% of the total — sits in a published unattributed bucket rather than being guessed into a seat. Corrections to the mapping are welcome.

Window: 1 August 2025 to 31 July 2026. Elexon revises settlement data for months; the weekly rebuild re-pulls the trailing 14 days.

Reconciliation against NESO

NESO publishes no wind-only curtailment cost in any machine-readable dataset, so the primary check is volumes, against NESO's per-BMU "Wind BOA Volumes" dataset, plus the GWh figures NESO quotes in its Monthly Balancing Cost Report. One unit note: NESO's BOA_Volume column is MW per half-hour settlement period, so energy = value × 0.5 — uncorrected it reads exactly double.

Spot checks against the report itself: May 2026 — ours 591 GWh vs NESO's quoted 591; June 2026 — ours 707 vs quoted 709. Per-BMU, 51 units above 20 GWh were compared; 1 fall outside ±2% and are listed in reconciliation.json. The build fails if any complete month disagrees by more than 2%.

Number 2 — consented but not built

Source: REPD quarterly extract (DESNZ), July 2026 edition. "Stuck" means development status Awaiting Construction or Under Construction with consent granted more than 24 months before the extract date (a 12-month cut is also computed and published). Consent date is the latest of planning permission granted, appeal granted, or Secretary of State granted. Refusals, withdrawals and the median months from application to decision use a five-year window.

Homes-equivalent = MW × technology load factor (DUKES 2026, five-year average) × 8,760 h ÷ 2,500 kWh (Ofgem's typical household electricity consumption, cut from 2,700 to 2,500 kWh in July 2026 — a change that quietly moved every homes-equivalent claim on the internet). REPD only includes projects above 150 kW (1 MW before 2021), understating small-scale totals. Statuses lag: where a "stuck" site demonstrably has a generating BMU we say so in the mapping notes.

Number 3 — the constraint bill per household

Source: NESO's Monthly Balancing Services Summary, constraints definition (England & Wales + Cheviot + Scotland constraint actions, constraint ancillary services, RoCoF/inertia, constrained sterilised headroom), cross-filled from the daily Constraint Breakdown dataset for months MBSS has not yet covered. Rolling 12 months: £2.5bn ÷ 27,292,462 GB households (Census 2021/22 via ONS, NRS and NISRA) = £92 per household. It is a national figure — the same for every GB seat; only the seat total varies with household count. Northern Ireland is in the all-Ireland SEM and is shown as not applicable, not as zero.

Relationship between the numbers: paying wind farms to switch off (£409m) is 16% of the whole constraint bill. The rest is mostly the mirror image — paying generators south of the bottleneck to turn up — plus voltage and stability actions. Number 1 is what was paid to wind here; Number 3 is the whole system bill.

Batteries

Battery figures come from the same planning database (REPD), Technology Type = Battery, bucketed as operating / being built / approved-not-built / refused (five-year window) / in planning. REPD carries no MWh (duration) column, so we never claim hours of storage. REPD's operating figure (5.0 GW) lags reality — industry trackers put the fleet at about 7.2 GW — so treat the operating number as a floor; the approved and refused numbers are planning records and are not affected by that lag.

The B6 bottleneck figures

The "wires are full" numbers on the front page come from NESO's day-ahead constraint flows and limits dataset for the SCOTEX (B6, Scotland–England) boundary: the flow is NESO's day-ahead modelled unconstrained flow — what wanted to cross — not metered outturn, because NESO publishes no outturn boundary flows. "Full" means wanted flow at or above the boundary limit.

Why a project is stuck, and what building it would do

Each project in a constituency briefing gets one reason, assigned by an ordered rule set — refused, then under appeal, then awaiting a planning decision past 12 months, then connections-reform outcome, then grid connection date, then under construction, then approved with no recorded progress, then in planning. First match wins; all matches are kept in the seat JSON. Where the public record does not say, the reason is "no public record of what is holding this up" — we never guess. "Who can move it" follows from the reason: the named planning authority, the Planning Inspectorate, NESO and the network owner, or the developer.

Connection dates come from NESO's Transmission Entry Capacity register and its Embedded register — the public queue of projects with agreed connection dates. NESO repeats rows per project stage, so we deduplicate on Project ID, keeping the earliest connection date. The registers carry no postcode or grid reference, so the join to the planning database is by normalised project name plus a capacity check, published as a mapping file with confidence grades and a candidate list. A guard rejects matches that rest only on a shared generic phrase — "Greener Grid Park" appears in many project names and is not evidence that two records are the same project. We would rather miss a match than assert a wrong one, so a project with no confident match simply shows its planning status instead of a connection date.

Connections reform (Gate 2). NESO does not publish project-level reform outcomes: its detailed results are aggregated by zone for commercial-sensitivity reasons. The machine-readable outcome is the Gate column added to the connection registers in November 2025, populated only as agreements are countersigned — so most of the queue is still blank, and a project without a Gate marking has not necessarily failed anything.

Impact estimates come from per-MW factors in a versioned file, each with a source, a date and a status. A sentence only appears when the factor for that technology exists — a missing factor means a missing sentence, never a zero. Three distinctions we keep visible, because they are where this kind of estimate usually goes wrong:

• In force vs proposed. The £5,000/MW/year community benefit rate for onshore wind is current government guidance. The solar and battery rates are from a Scottish Government consultation paper and are not in force; where a seat's figure uses them, the page says so.
• Published vs derived. The VOA publishes no per-MW business rates factor for wind or storage — it values them on a receipts-and-expenditure basis with unpublished models. Our figures are derived from the VOA's own 2026 rating list joined to project capacities, and are described on the page as what comparable sites are rated at, not as an official rate. Solar is the exception: there is a published VOA / Solar Energy UK agreement.
• Workers on site, not job-years. Construction figures are peak concurrent workers, taken from Scottish Government workforce research and consented NSIP environmental statements. Published UK job-year estimates vary by a factor of two to four depending on whether indirect effects are counted, so we do not use them. No credible per-MW construction figure exists for grid-scale batteries, so batteries are left out of that line entirely.

What we cannot see

• Distribution-level curtailment is invisible. Wind farms connected to local networks are curtailed under Active Network Management by network operators; those payments never appear in national balancing data. English and Welsh seats showing "not visible" are not necessarily curtailment-free — the national record simply cannot see them. This means the site understates the true cost.
• Aggregated portfolio BMUs (e.g. supplier units bundling several small farms) cannot be honestly attributed to one site and stay in the unattributed bucket.
• REPD status lag: planning statuses can run a quarter behind reality.

Comparators

Every tangible framing resolves through a versioned comparators file; nothing is hardcoded. Values, sources and as-of dates:

Changelog

• 2026-08-25 — first build: REPD Q2 2026, Elexon settlement stack Aug 2025–Jul 2026, MBSS to Jun 2026 (Jul 2026 provisional from Constraint Breakdown), Census 2021/22 households, comparators verified against primary sources.