How the queue data is built, and where it stops being a fact

The rule throughout is that data which is absent is never invented. Where something is inferred it is labelled, drawn differently, and explained at the point of use.

The register

One row per project identifier. Capacity is the sum of the register's own MW increase or decrease, which is new capacity only. Cumulative total capacity is deliberately not used, because it equals connected plus increase and therefore counts plant that is already standing at a brownfield site. Using it would inflate the pipeline from 426.7 GW to 438.3 GW.

Coordinates, and four grades of them

The register names a connection site on every row and publishes no coordinate for any of them. Four routes, in descending confidence:

A fragment of a name is never a match. An early version placed six unrelated sites at one coordinate because it matched on the word bess after stripping farm, and another matched Botley West to Botley Wood. Both rules were deleted rather than tuned. A multi word name has to find a multi word match, and a directional qualifier the candidate does not share is a rejection.

The connection point score

Four components with stated maxima, then a deliverability gate. Contested out of 40 is all storage capacity seeking the point, against the largest stack in the country. Returning on evidence out of 30 is Gate 2 capacity whose contracted date has already passed. Returning on risk out of 20 is Gate 2 capacity scoring high on delivery risk but not yet late. Persistence out of 10 asks whether more than one applicant is present, so churn is not a single counterparty event. The total is multiplied by 1.0 where storage is already energised at the point, 0.8 where Gate 2 is present and consented, 0.5 where Gate 2 is present with nothing consented, and 0.25 where there is no Gate 2 position to take.

Delivery risk

Consent gap out of 50, days past the contracted date out of 30, capacity scale out of 20. The lead times behind the consent gap, 1,095 days from scoping, 730 from awaiting consents, 547 from consented and 180 from under construction, are inference and are published nowhere. They are stated so you can disagree with them.

Ownership

Applicants are grouped into platforms by shared registered address at Companies House, which is strong but circumstantial evidence of common control, and it is graded. An address only groups where fewer than five hundred companies are registered at that postcode. Fifteen postcodes in this set host more than a thousand companies each and one hosts over seventy thousand, so a shared address there means nothing and those applicants are left ungrouped.

Balancing mechanism revenue

Every accepted bid and offer for 364 days, taken from Elexon's settlement stack, aggregated by unit. Validated against NESO's published balancing cost at a ratio of 0.97 for a sample three days before any figure was used. Offer revenue is reported because it is unambiguous. Net balancing cash is not profit and is not presented as such.

What is refreshed, and when

The connection registers are read on Tuesdays and Fridays and compared with the previous capture, so entries, exits, gate moves, capacity changes and applicant changes are recorded rather than reconstructed. The capacity market, planning and Companies House sets refresh on their publishers' own cycles, which the tool states next to each.

Every source, when it was last read, and under what licence

This table used to live inside the tool, where it reappeared in a 350 pixel column beside a map that has nothing to do with it. It belongs here, where it can be read once.

All methods