GB Connection Slippage

Every generator, battery or interconnector that wants to plug into Britain's high-voltage grid holds a contract with a target connection date, and the grid operator publishes the list twice a week, overwriting the last copy. This page keeps the copies. For projects on the list at both ends of a period, it adds up how far each target date moved, weighted by the project's capacity: a 100 MW project whose date moves one year later counts 100 megawatt-years. Projects that joined, left or changed size in the period are listed beside the total, never inside it. The series is built from 694 copies of the register, 31 Jan 2014 to 22 Jul 2025, plus 4 copies since 19 May 2026 under the reformed regime.

Correction, 30 Sep 2026. The megawatt-years on this page are gross of cause: the register never prints why a date moved, and investigation 006, over the 213 slips of 24 months or more among the 863 project-stages it could join to the Transmission Works Register, classed 60% as project-led, 26% as works-led (a same-direction move in an enabling scheme's proxy date inside the window) and 14% as unattributable (investigations/006-connection-slippage-attribution/RESULTS.md and evidence/twr-attribution-summary.json); no figure on this page changes.

Between 19 Jul 2024 and 22 Jul 2025, 1,147 project-stages were on the register at both dates and 889 of them carried a connection date on both copies. +53,221 MW-years is movement the register's own stage labels determine. 88 project groups that were split, merged, renumbered or printed repeated stages, at either end or in any copy between, or that went missing from a copy between, add +4,602 under version 2's rule and +8,322 under content matching, so the headline is +57,823 or +61,543 depending on the rule; other pairings are possible and are not bounded here. Chained copy to copy across the old regime only (694 copies, 31 Jan 2014 to 22 Jul 2025; the reformed regime is a separate series), the movement the stage labels determine is +392,126 megawatt-years; with the rest it is +532,780 under version 2's rule (26.4% undetermined) or +534,181 under content matching (26.6% undetermined). Under version 2's rule the chain exceeds the sum of the year windows below (+293,599) because it also counts project-stages present in two consecutive copies but not at both ends of a year.

Where a project prints several rows, the register does not always say which row of one copy is which row of another: when its stages were split, merged or renumbered, printed the same stage twice, or went missing from a copy in between. Movement in every other project is determined by the register's own stage labels, and every identity rule that pairs a stage with the stage of the same label gives the same figure. For the rest this page gives the figure under two named rules: version 2's rule (rows numbered by date) and content matching (rows followed by their capacity and date). The two are not bounds; other pairings are possible.

Year by year

Each row compares the first copy of the register in one year with the first copy of the next. Project-stages at both dates is the population the movement is summed over; a project that was renamed, changed customer or changed its connection site in between counts as having left and joined, never as moved. Dated at both ends is how many of those carried a connection date on both copies; only they can move, so the later, earlier and unchanged counts add up to it. Joined and left show the count and the capacity in megawatts. A positive movement is later; megawatt-years are rounded to whole numbers here and kept to three decimals in the evidence file. The last column marks a comparison whose population at both dates is under half of the earlier copy's rows (the declaration's F2): the figure stands, and the reader is told it rests on a thin population. The 2020 and 2022 windows straddle the register's own format changes (a new column set from June 2020; connection-site names respelled from December 2022), which the earlier investigation found break the identity of many rows; F2 there reflects the register, not the projects.

In 2018, 2020 and 2023 the movement the register's stage labels determine is under half of the year's figure under at least one rule: 2018, +26,714 of +68,399 under version 2's rule and +68,220 under content matching; 2020, +1,371 of +2,903 under version 2's rule and +2,903 under content matching; 2023, +1,770 of +13,584 under version 2's rule and +13,584 under content matching. In those years more than half of that figure comes from project groups whose pairing the register does not determine.

Since the reform: a separate series

From December 2025 the register carries dates re-baselined under the grid operator's connections reform (Gate 2 offers). Movement across that change is the reform, not slippage, so these copies form their own series, compared copy to copy from 19 May 2026, and are never added to the figures above.

Propositions, declared before the run

• P1: every complete calendar-year window of the old regime has positive net movement: holds over 11 complete years (2014 to 2024).
• P2: chained net movement of the reformed regime from its first copy to the first copy at least 365 days later is positive: undecided (falsifier date 30 Jun 2027).
• F1 fired on 6 link(s) between consecutive copies; F2 fired on 192 year-on-year row(s). Both are marked in the tables.

Gaps, breaks and copies not used

• Regime break. No copy is held between 22 Jul 2025 and 19 May 2026 (301 days): the grid operator's connections-reform re-baselining. Copies from December 2025 onward are a separate series below and are never chained to the earlier one.
• No copy held between 4 Jul 2014 and 7 Oct 2014 (95 days).
• No copy held between 1 Apr 2025 and 1 Jul 2025 (91 days).
• The copy of 29 Jan 2021 carried dates with day and month exchanged; it is read with them exchanged back.
• The copy of 19 Feb 2021 carried dates with day and month exchanged; it is read with them exchanged back.
• The copy of 11 Mar 2022 carried dates with day and month exchanged; it is read with them exchanged back.
• No copy held between 19 May 2026 and 22 Aug 2026 (95 days).
• The copy of 8 May 2015 is a suspect copy: 225 rows against 413 in the previous copy (46% fewer); the next copy recovered, so it is left out of the series.
• The copy of 28 Nov 2023 is a suspect copy: 673 rows against 1392 in the previous copy (52% fewer); the next copy recovered, so it is left out of the series.
• 32 held copies could not be parsed and are not in the series (24 Feb 2014, 28 Feb 2014, 4 Mar 2014, 11 Mar 2014, 20 Mar 2014, 25 Mar 2014, …).

How the numbers are made

The rules were written down and sealed before the series was computed, and the sealed text is never edited afterwards. The list is the grid operator's TEC Register (Transmission Entry Capacity), assembled from its own disclosures under the Environmental Information Regulations, Wayback Machine captures and the live data portal. A project-stage is one row of the register: the project's name, customer and connection site together with its stage. Its capacity is the stage's MW increase as published, and its date is the register's effective-from date for that capacity. Movement is the difference between the two dates in days divided by 365.25, times the capacity in the earlier copy, so a change of capacity never enters the movement; it is reported beside it. Rows without a parseable date on both sides, or without a capacity, are counted and contribute nothing. Nothing here says why a date moved, whether a project will connect, or what any party is entitled to; the grid operator's own caveat that project status is its best-known classification applies throughout.

Declaration (SHA-256 7d07cbd159ac14d2…): investigations/014-gb-connection-slippage/DECLARATION-v4.md, witnessed by OpenTimestamps and two RFC 3161 authorities before the series was computed. Evidence, every copy's digest and every row: evidence/v4/. Code and tests: github.com/jordan-dimov/grid-mysteries. Figures computed 2026-09-23 and read from evidence/v4/series.json; the page is a pure function of the evidence and of corrections.json. The two tables of individual copies show the last 400 days; every copy since 2014 is one line of evidence/v4/rows.ndjson, appended and never rewritten.

Corrections

1. 24 Sep 2026: The introduction said the series is built from 700 copies of the register, 31 Jan 2014 to 22 Jul 2025, plus 4 copies since 19 May 2026. 700 was every usable copy, so it counted the 4 reformed-regime copies twice and the 2 copies left out by the partial-export rule as used. The old-regime series is built from 694 copies. No movement figure changed.
2. 30 Sep 2026: The megawatt-years on this page are gross of cause: the register never prints why a date moved, and investigation 006, over the 213 slips of 24 months or more among the 863 project-stages it could join to the Transmission Works Register, classed 60% as project-led, 26% as works-led (a same-direction move in an enabling scheme's proxy date inside the window) and 14% as unattributable (investigations/006-connection-slippage-attribution/RESULTS.md and evidence/twr-attribution-summary.json); no figure on this page changes.