This window — measured, not modelled

• Frequency response remained the largest measured stream (65% vs BM 18% in 2026Q2).
• Active fleet grew 15% inside the window — 5.5 → 6.4 GW (130 → 153 BMUs). Supply growth is the denominator behind every per-MW move.
• 2h-class cycling reached 1.07 cycles/day in 2026Q2 — the highest of the window.
• Top-quartile units out-earned the median by £35.2k/MW·yr (median £35.1k, top-quartile mean £70.2k; n=113).

The rotation — £/MW·yr by stream, per quarter

Per-quarter annualised revenue per active MW. Frequency response shrinks as the fleet grows and FR prices saturate; the Balancing Mechanism absorbs the MW. Capacity Market is an annual stream (dashed window-level line) — it has no quarterly settlement basis. The former combined "Frequency response" line is shown split into response (DC/DM/DR) and reserve (QR/BR/SR) — a presentation change registered 2026-07-23; the two segments stack to exactly the published total.

The floor, re-measured — companion series for the 2026Q3 re-basing

Two of the pre-announced 2026Q3 basis changes (page 5) are measured here as companions; the published floor above is unchanged until the window advance, and nothing is restated. BOA-netted removes accepted Balancing Mechanism energy from metered volume before MID-marking, so the wholesale and BM lines stop double-marking the same MWh: in 2026Q2, accepted offers were 41% of fleet discharge and accepted bids 55% of fleet charge, and netting them moves that quarter's floor by +£20.1M (the gross line UNDERSTATES that quarter); across the window the netted floor sits +£21.7M from gross. + TLM then credits metered energy the way settlement does — Transmission Loss Multipliers per settlement period, delivering on discharge and offtaking on charge, each unit's zonal loss factor entering as a class-weighted mean. Applied to a net floor whose gross legs dwarf it, the measured TLM effect is −£5.8M (−51% of the trailing floor); the composed series ends +£15.9M from gross. Where a companion line runs negative, that is the measurement, printed. Sources: Elexon BOAV settlement acceptance volumes and SAA settlement-run TLM data (the run used per day, per-quarter series and gross-line reconciliation ship in the evidence JSON and the data pack).

The measured per-unit distribution

Each dot is one BMU's trailing-year measured revenue per MW under the index conventions (wholesale floorᶠ + EBOCF net BM + EAC FR; registry-MW denominator). Bars mark P25/median/P75; the median is copper. Units active in all four quarters only (survivor cohort: 122 of 168 registry BMUs; 9 excluded on the disclosed capacity-plausibility screens — see page 4); 7 units have negative measured totals (real, convention-driven: BM-instructed charging pays out in the wholesale leg). Spread partly reflects strategy and location, not only operator skill. Classes with too few screened units for stable percentiles (4h) are not shown. Per-unit totals are NOT individually "≥" floors: with no BOA netting, offer-heavy units can be overstated and bid-/response-heavy units understated — the floor direction holds at fleet level only. This strip orders 2h ahead of 1h on per-unit medians while the class table below orders 1h ahead on MW-weighted totals — its premium sits in the FR leg: the 2h class carries a heavy left tail (curtailment-zone units, some negative) that drags its MW-weighted mean below its median. Both are correct; they answer different questions.

Strategy mix — the measured split by operating strategy

Classification registered 2026-07-24, bands frozen: per behaviour-gated unit over the trailing window, FR share = frequency-response £ over (FR £ + BM £ + wholesale-floorᶠ £ clamped at zero); FR-annuity at share ≥ 0.60, Trader at share ≤ 0.25, Mixed otherwise. Informational, never graded; the bands change by a new registration only. Measured this window: the 70 FR-annuity units (43.7% of gated MW) measure a £-0.1kᶠ/MW·yr wholesale floor beside £29.5k/MW·yr of frequency response. The 46 Trader units measure £18.3k/MW·yr in the Balancing Mechanism and £3.8kᶠ/MW·yr on the wholesale floor, with £4.4k/MW·yr of frequency response. The 35 Mixed units measure £13.9k/MW·yr FR, £20.2k/MW·yr BM and £4.9kᶠ/MW·yr wholesale floor — £39.0k/MW·yr combined. The full per-asset dataset behind this panel (151 gated units: strategy class, stream split, within-class percentile, cycling, screens; gate failures in an exclusions disclosure) is published beside the index. In plain terms: 44% of gated battery capacity is run mainly for frequency-response income, 34% mainly trades energy through the Balancing Mechanism and wholesale, and the remaining 22% does both.

Location: the TNUoS-zone BM split

Net BM = bid + offer EBOCF cashflows: curtailment-absorbing zones (Scottish boundary) are net NEGATIVE because BM-instructed cheap charging is paid-for energy whose value lands in the wholesale leg, not a loss; the OFFER leg (discharge instructions only) is shown alongside. GB mean net BM £12.2k/MW·yr. Postcode area → zone is an approximate hand mapping. Greyed rows have n<3 — treat as indicative. 29 units (887 MW·yr) have no mappable zone and sit in the GB mean only. The best-to-worst spread is £85.7k/MW·yr — siting moves BM revenue more than duration does.

How each stream is measured

What we exclude, and why (quantified)

Known measurement limitations — measured

One entry per limitation: what · measured magnitude · consequence for the reader · status. Magnitudes are from the adversarial audit dated 2026-07-23, run against the pinned 2025Q3–2026Q2 caches and independently verified; they are stated two-sided where the exposure runs both ways. No entry changes this window's published basis — basis changes are pre-announced in the dated box below, number-changing corrections are collected in the dated record, and none applies retroactively.

Pre-registered companions — measured beside the published basis

Registered 2026-07-23; every basis change applies at the 2026Q3 window advance only and will be dual-published against the outgoing basis for at least four quarters. Nothing on this page changes a published figure — each companion is measured NOW so the 2025Q3-2026Q2 record carries both readings. The frequency-response/reserve split from the same registration is APPLIED throughout this report (presentation and naming only; every split pair sums to the published total exactly).

CPI-indexed Capacity Market companion

CM clearing prices indexed from auction-year to delivery-year money per the Capacity Market's own mechanism (Electricity Capacity Regulations 2014 Sch. 1: winter-average (Oct–Apr) CPI ratio; T-4/T-3 prices only — T-1 clearing prices are not CPI-adjusted, which is why the all-T-1 4h class moves +0.0%). ONS series: CPI all items, CDID D7BT, dataset MM23 (2015=100); the Regulations name "CPI" without an ONS series id, so D7BT is the documented choice. Factors this window: 2024-25 T-4 ×1.2110; 2025-26 T-4 ×1.1736; the statutory base period is set per auction — the winter ending in the auction-held year is a disclosed approximation. Measured: measured fleet uplift +9.6% (£2,187 → £2,397/MW·yr) — inside the audited ~+8–9% expectation band (£2,362–2,406/MW·yr): the Rules' mechanism indexes T-4/T-3 prices only, on winter-average CPI, so the all-T-1 4h class stays at +0.0% while the T-4-heavy 1h class runs above the fleet mean. † withheld on the published basis (partial-window class); the entrant-aligned companion below states it consistently.

Entrant-aligned (prorated) denominator companion

Companion convention: active-MW-days run continuously from each unit's first metered-volume quarter (commissioning) to the window end — idle quarters count as in-denominator downtime — and CM accrual realigns to the same entrant dates (numerators unchanged). First-metered entrants: 2025Q3: 13 units / 739 MW (13.4%) · 2025Q4: 14 units / 545 MW (9.0%) · 2026Q1: 9 units / 288 MW (4.6%) · 2026Q2: 7 units / 256 MW (4.0%); re-entries after an idle quarter (returns, not commissioning: 57 MW in 2025Q3, 57 MW in 2026Q2) are shown separately from entrants (corrections record, page 5). Resolution: sub-quarter first-metered dates are not derivable from the pinned per-BMU quarter caches, so this companion realigns at quarter resolution; the day-level proration lands with the 2026Q3 refetch. The realigned CM column gives the 4h class the consistent per-MW/yr the published basis withholds (numerator and denominator now cover the same days).

Traded-spread companion — market-comparable context

Explicitly not a settlement measurement. Per class: the mean daily top-Nh-minus-bottom-Nh MID spread (N = class duration hours; MID both providers, volume-weighted) × the measured class spread-capture rate, annualised at one cycle/day (× N MWh per MW per day × 365) — published beside the settlement floor as market-comparable context. The gap between this line and the floorᶠ is the measured size of the metered-at-MID convention, not extra revenue; the capture rate's available-spread convention (daily max−min period) is disclosed in the evidence JSON.

2026Q3 preview — the re-based cut (companion)

Pre-registered 2026-07-23, before any data was fetched for it; applies at the 2026Q3 window advance only, never retroactively; dual-published beside the outgoing basis for at least four quarters. Everything on this page is COMPANION machinery — the graded basis (the published index, its targets and its grades) is unchanged. Two rules act here: a behavioural battery gate (a unit enters companion numerators and MW-day denominators only if its trailing discharge/charge ratio sits in 0.50–1.02; the two-sided sensitivity is on page 5) and metered-implied duration (the p99.5 contiguous-discharge run, page 4) as the primary classification. Gate failures are unclassed — hybrid-suspect units never enter a class row.

Class economics, re-based vs current (per MW·yr)

Same streams, same conventions, same pinned settlement data — only membership (gate + widening) and classification move. n = latest-quarter unit count. † CM per-MW/yr withheld for partial-window classes (same refusal rule as the current basis). The implied cut brings 64 previously un-banded gated-in units (2,886 MW) into the class split, and moves 23 CM-banded units (975 MW) to a different class than their registration. Like-for-like continuity (reclassifying only the CM-banded universe, before the gate): 1h and 2h totals land 1.8% apart (≥ £31.3k vs ≥ £31.9k/MW·yr) — the within-2% fact from page 4, reproduced by this build. Gated fleet, three measured streams: ≥ £32.0k/MW·yr vs ≥ £31.5k on the current basis.