Conventions

Per-MW figures are active-MW-day weighted on the re-based basis applied at 2025Q4-2026Q3: numerators and denominators cover exactly the units in each quarter's day-prorated denominator — a unit counts full days in every quarter it metered volume and none in a quarter it did not (the outgoing active-quarter scoping, retained); a commissioning entrant's first quarter counts from its first metered-volume day; a unit metered before the window is never an entrant and its active quarters count in full — and CM accrues from each site's entrant date to the window end (a site keeps accruing through a quarter its unit sat idle; the exit-side mismatch is disclosed, not bridged). The wholesale line is a metered-at-MID floorᶠ, TLM-adjusted and net of accepted Balancing Mechanism energy and of ABSVD response energy at MID (the outgoing gross series is dual-published — page-1 table and page 6). Membership is the registry widened to every Elexon-registered battery BMU with metered volume, each unit subject to the behavioural gate (page 7). Cycles divide metered discharge by MW × CM-register duration hours, grouped by the metered-implied class. One registry, composed as of the build date. Initial settlement for the newest quarter; older quarters are pinned at their fetch vintage (restatement policy — page 5); figures revise. Public cross-check: the measured 2h cycles (0.94/day in 2026Q3) sit BELOW the public 1.2–1.5 band. We publish the measurement, not the band. Full conventions, every netting term and the complete outgoing basis ship in the evidence JSON (/api/index) and the data pack.

Duration-class provenance. Classes are metered-implied (applied at 2025Q4-2026Q3): per unit, the p99.5 quantile of rolling 1–16-settlement-period discharge-MWh sums over the window, divided by capacity, through the standard edges (<1.5h → 1h, <3h → 2h, <6h → 4h, else 8h); gate failures are unclassed. The CM-register band cut — the outgoing convention, in which each site took the band carrying the most registered MW across its capacity-market components — is dual-published in the evidence JSON and compared on page 7. The register records what was registered, which need not be how a unit runs: on this window (2025Q4-2026Q3, the registered pooled quantile) the metered-implied reading found 8 of 40 "1h"-banded units (356 MW, 23% of the class) sustaining ≥1.5h, and every unit of the banded "4h" class (6 units, 313 MW) behaved as 2h; units with no band at all (the July audit counted 70 active units, 2,993 MW, ~45% of active fleet MW) now enter the class split. On the July audit's window (2025Q3–2026Q2) reclassifying the banded universe left 1h and 2h within 2% of each other; on this window the same like-for-like cut puts them 25% apart (1h ≥ £33.3k vs 2h ≥ £25.0k/MW·yr incl. CM). In plain terms: a battery's duration class comes from how it actually ran over the year, and the class on its paperwork is kept beside it for continuity.

ABSVD treatment. Measured in both bases. Energy delivered by primary BM Units under NESO frequency-response contracts is removed from imbalance settlement via ABSVD (deemed volumes; NESO's C16 methodology), so that energy is neither bought nor sold at market prices, while the outgoing wholesale floorᶠ marked ALL B1610 metered volume at MID — penalising response-heavy units; the published floorᶠ nets that energy out (17.10). Per-BMU Applicable Balancing Services Volumes are freely published on Elexon's Insights API (dataset QAS, per settlement period, history to 2022; access probe dated 2026-07-23 — corrections record, page 5). Netting is INSIDE the published floorᶠ since the 2025Q4-2026Q3 re-basing (17.10 applied): on the published universe the un-netted and netted readings are fleet £14.5k → £16.9k; 1h £1.7k → −£6.5k; 2h £23.2k → £32.6k; 4h −£5.1k → −£11.0k/MW·yr (un-netted → netted). On the published universe netting does not raise the short-duration class floors; deemed response volumes can exceed a unit's metered discharge, which is why a response-heavy class's netted floor can read negative. The audit's FR-heavy subset, reproduced on the OUTGOING basis (outgoing universe, gross floor — an audit reproduction, not a published figure): 88 units, 3.4 GW, netting lowers it from −£1.7k to −£9.8k/MW·yr (46 → 35 units with a negative floor). The mechanism, measured: that cohort's ABSVD is net DELIVERED response which the un-netted floor had been crediting at MID; charge-side response dominates the longer classes, where netting hands the credit back. The un-netted basis is dual-published (the outgoing basis in the evidence JSON; the QAS term per class on page 6). FR revenue itself is EAC contracted results, unreachable by settlement adjustments; BM is accepted-BOA cashflows, which ABSVD does not adjust. Secondary and non-BM units receive no frequency-response ABSVD and largely sit outside the B1610 universe — a disclosed composition limit (BSC Issue 114 / P502 are live on exactly this). In plain terms: the outgoing floor convention priced energy that settlement never traded, which penalised batteries that mostly sell response; the published floor removes that energy, and the old reading is printed beside it above.

Availability basis (per-MW denominators). The active-MW-day denominator is day-prorated at entry and quarter-scoped thereafter: a commissioning unit counts from its first metered-volume settlement date, and every unit counts full days in each quarter it metered volume and none in a quarter it did not, so outage days inside an active quarter remain in the denominator while a whole idle quarter after entry carries no MW-days; per-MW figures are net of in-quarter downtime only (its site's CM keeps accruing to the window end — the exit-side mismatch, disclosed). The survivor cohort on page 2 (135 of 168 registry BMUs) is the fixed-composition year-on-year cut. In plain terms: a new battery counts from its first metered day; a battery that ran at all in a quarter counts at full size for that quarter, and one that sat idle for a whole quarter drops out of it. Downtime inside a live quarter is in the averages; whole idle quarters are not.

Bridge to the benchmark conventions — measured leg by leg

The headline above is a public-data floor basis. Each row below is a measured quantity from our own open inputs, applied in the fixed order shown on the published re-based cohort (2025Q4-2026Q3); legs we cannot measure from open data are named and left in the residual — none is a plug, and no leg is derived from a benchmark figure. Benchmark: Modo Energy's ME BESS GB, the FCA-regulated benchmark; its conventions are taken from its public methodology page (version 3.1, last updated 2026-05-18, read 2026-09-15). Consequence: the bridged totals are the closest like-for-like reading of this record under the benchmark's conventions that open data supports; what remains is named and left unestimated. Status: the benchmark column is off; with it off this page prints no benchmark figure and no ratio.

Leg 1 enters the incl-CM total only; the ≥6 MW screen also re-scales the CM line, and that CM-side effect sits inside the incl-CM total (carried per leg in the artifact). BOA and ABSVD netting, entrant proration and CPI indexation are inside the published figure for this window and are not applied again. Reserve split as published: fleet response £14,358 / reserve £2,690 per MW·yr. Balancing Reserve rows are present in every quarter of this window.

Informational companions — applied in the published figure from the re-basing

2H cut, published beside the 2h class (never mixed). The CM-declared 1.5 ≤ h < 2.5 cohort (41 units, 1,676 MW active-MW-day weighted) on the same waterfall: ex-CM £59,771 → bridged ex-CM £59,771, incl-CM £64,441/MW·yr. Duration bases side by side: published 2h class = 1.5–<3.0 h metered-implied (CE's own edges); CM-banded 1.5–<3.0 h = the dual-published outgoing cut; this cut = 1.5–<2.5 h (CM register); the benchmark's = rated energy/power per its methodology page (§4.2.2), a field CE's open inputs do not carry. No public 2H print exists for this window.

Where our convention understates (measured legs that move the fleet figure up): Capacity Market inside the total. Where it overstates (measured legs that move it down): none.

Named residual components — what remains after the measured legs, with the input each needs; no magnitude is assigned: Wholesale price basis — the benchmark's N2EX/EPEX day-ahead mean and RPD half-hourly marks vs the N2EX GB day-ahead hourly price (licensed-only remainder) (needs licensed EPEX day-ahead and RPD half-hourly price series (not open data) — not computed); Rated power vs registered export limit / max executed MW in the divisor (needs rated power per asset (no open field)); 'No shared meter' screen and the FPN-or-contract daily activity rule (needs a co-location field and per-day FPN submissions); Battery auction revenue outside the BMU perimeter (bracket on the residual row) (needs a unit-to-site mapping for alias and aggregator auction units); Availability/performance scaling and the EFA-day vs settlement-day edge (needs per-unit performance-monitoring outcomes); Settlement-run vintage of the pinned data vs the benchmark's revision ladder (needs docs/evidence/bess_index_restatement.json (first committed at the 2026Q3 advance)); Universe composition (the benchmark's asset list vs the CE registry as-of build date) (needs the benchmark's asset list (not published)); LCCC monthly payment weighting of CM within the window (needs the public LCCC monthly weighting profile applied per delivery year (v2 table)); Capacity Market obligations not attached by the postcode join (coverage bracket on the residual row) (needs a CMU-to-BMU join beyond postcode); Duration basis — metered-implied classes (published from the re-basing) vs the benchmark's rated energy over rated power (needs rated MWh per asset (no open field)). Brackets on the residual row of the artifact: battery auction revenue outside the BMU perimeter, and the share of register storage-CMU obligations the postcode join attaches.

History coverage. reserve split measured from 2023Q4 (FR not measured 2022Q1–2023Q3; 2023Q4 partial); BOA netting from 2025Q1; ABSVD netting from 2025Q3; ≥6 MW screen every quarter from 2022Q1; Capacity Market on the trailing window only; BOA and ABSVD netting, proration and CPI indexation are inside the published figure from 2025Q4-2026Q3 (earlier rows keep the conventions they were published on). Balancing Reserve: 2024Q2, 2024Q3, 2024Q4, 2025Q1, 2025Q2, 2025Q3 carry 'excludes Balancing Reserve — NESO EAC-BR results dataset not fetched'. Pre-EAC rows carry 'FR not measured' and no benchmark delta.

Corrections & methodology changes — dated record

2026-07-23 · out-of-perimeter FR corrected £31.3k → £1.71m (2026Q2; the prior print carried the alias slice only — a 55× understatement) · page-3 figures restated; per-MW figures unaffected

2026-07-23 · per-BMU ABSVD access re-classified: the "S0142 settlement reports via the P114 service only" claim was measured WRONG by a dated access probe (2026-07-23) — Elexon's Insights API publishes dataset QAS freely · ABSVD-netted companion published (page 4)

2026-07-23 · CM attachment caveat corrected — one pumped-storage station is 65% of the prior "1,295 MW cannot be attached" figure · battery-side skip was overstated ~2.8×

2026-07-23 · BM/wholesale overlap re-classified from marginal to first-order and quantified · 2026Q2 net −£11.4M; measured as the page-1 companion

2026-07-24 · the 2h-class floor guidance withdrawn (measured 41% inflated by non-battery units) · class-level floors carry no "≥" guarantee

2026-07-24 · 2026Q2 entrant share corrected: 4.9% counted a post-idle return as an entrant · returns shown separately from entrants (page 6)

2026-09-15 · the hand-typed benchmark-context paragraph (its ratio and two magnitudes were not reproducible from any committed artifact) retired · replaced by the computed bridge section above, sourced only from the reconciliation artifact; no benchmark figure or ratio is printed while the benchmark column is off

2026-09-28 · the metered-implied duration computation corrected to the registered wording — one p99.5 quantile over the window's pooled contiguous-discharge sums, not the per-quarter maximum the July preview used · 9 units / 322 MW change class in the page-7 cut (eight 2h → 1h, one 4h → 2h); the "1h"-band finding on page 4, on the same ungated CM-banded universe, reads 8 of 40 units (356 MW) where the July print read 16 of 40 (606 MW) — the computation changed, not the universe; no published figure moves

2025Q4-2026Q3 · the re-basing registered 2026-07-23 APPLIED at the window advance — metered-implied classes, the behavioural gate, CPI-indexed CM, TLM, day proration with realigned CM accrual, widened membership and BOA/ABSVD netting are inside the published figures (box below) · the outgoing basis dual-published; no history row moves

2026-10-10 · methods release 2026-10: the headline wholesale line re-published as the trading position with the measured floor beside it (protocol §39; window state full, headline the trading-position line) · widened members' frequency response counted on the registry's own key, the uncounted convention published beside it (§40) · the Capacity Market accrual line in the living series (§41) · the 2026-10-09 record carried verbatim under vintage_2026_10_09 · no pinned quarter restated

2026-10-12 · plain-language edition of this report, its landing page and the living index page: presentation and wording only; the trading-position headline and the measured floor beside it, every registered label and footnote, every table and every figure unchanged; no pinned quarter restated; the 2026-10-11 render kept as a frozen archive beside the 2026-10-09 render

2026-10-13 · the "auction units we cannot map to any BMU" figure on pages 3 and 5 split: £6.4m of the £21.0m it printed for the window (£2.77m of £7.41m in 2026Q3) was frequency-response revenue at the BMUs the behavioural gate excluded, now named beside it; the two parts add to the figure printed before; no per-MW figure moves

No correction applies retroactively to a pinned quarter. The full trail is in the CECadence changelog; everything else in this report states the current convention.

Applied at the 2025Q4-2026Q3 re-basing — registered 2026-07-23

The number-changing corrections pre-announced on 2026-07-23, now APPLIED to the published basis, with the outgoing basis dual-published for continuity (at least four quarters; the CPI and proration alongside items for one year): (1) primary duration classification is metered-implied duration, the CM-band cut published beside it (page 7); (2) the behavioural battery gate is on the registry, the excluded units and £ published as a disclosed exclusion (page 7); (3) CM clearing prices are CPI-indexed to delivery year (page 6); (4) Transmission Loss Multipliers are applied to marked volumes (page-1 table); (5) MW-days are prorated from each unit's first metered day and CM accrual is aligned to the same entrant dates (page 6); (6) registry membership is widened beyond the EAC-participation condition (page 7); (7) ABSVD deemed-volume netting is inside the floor — its access condition was RESOLVED by a dated probe (2026-07-23): per-BMU QAS volumes are freely published on Elexon's Insights API (the S0142/P114 settlement-report route is not needed for volumes); (8) the traded-spread wholesale companion line is published alongside the floor (pages 1 and 6). Every change to the graded basis went through pre-registration before application; none is applied retroactively — history rows and earlier windows stay on the outgoing basis; all are dual-published for continuity. The response/reserve split from the same registration is carried throughout this report — presentation and naming only, totals unchanged.

Method updated 2026-10 (trading-position wholesale); quarters on the previous method only: 2022Q2–Q4, 2023Q1, 2023Q3–Q4 (pn / sysprice not complete in Elexon's served data)

Never graded, never a target, never a vintage record, never a restatement of any pinned record; recomputed in full from the immutable pins at any later rebuild. No quarter of this backcast enters the graded record.