As more storage is built, batteries compete away the very price spreads they earn from. A naive "price-taker" forecast — which assumes prices are unaffected by the fleet — overstates 2030 BESS revenue by 100–300%+. CECadence solves prices, fleet build-out and dispatch jointly to convergence, and reports the bias — naive minus equilibrium — as a headline number, because that bias is what decides a project's DSCR.
On a 100 MW / 2-hour project (~£50–100M capex), pricing the build on a naive forecast is tens of millions of mispriced value and a covenant breach a few years in. Most of the market still forecasts BESS revenue as a battery dispatching against an exogenous price curve. That assumption silently breaks at fleet scale — and breaks worst exactly where the money is, around NESO's 2030 trajectory.
The fixed point — fleet ↔ price ↔ dispatch coupled and solved jointly by Krasnoselski–Mann iteration — recovers the answer a price-taker model cannot. Stylised; the in-app cannibalisation sweep is computed per scenario-year.
Everything you do is one of two things: run a whole-GB scenario, or appraise a candidate site against that equilibrium. A candidate joins the fixed point as an extra asset and is valued at the with-candidate equilibrium — never against exogenous prices.
Pick a scenario and year; CECadence solves the cannibalisation fixed point and returns the equilibrium price duration curve, the naive curve, the bias, and a per-duration revenue stack (1h / 2h / 4h) — wholesale arbitrage, frequency response, Capacity Market and Balancing Mechanism, in £/MW/yr. Sensitivities, Monte Carlo bands and the cannibalisation sweep are one click away.
Enter a site — power, duration, commission year, TNUoS zone, capex/opex, round-trip efficiency. Get equilibrium revenue, the cannibalisation externality on the existing fleet, project finance (NPV, IRR, DSCR), an NPV tornado, and an indicative CEAtlas grid-connection estimate. Save sites to a portfolio and compare 2–4 head-to-head.
Wholesale arbitrage is the contested one — and the only line where cannibalisation bites. The ancillary stack is co-modelled with participation factors so the products don't double-count.
Per-cohort dispatch LP — SoC balance, round-trip efficiency, degradation as a marginal cost, least-cycling tie-break — solved against the equilibrium price, with the naive price-taker number alongside.
Saturation-priced reserve revenue calibrated to the NESO EAC trailing year, with an optional product-faithful FR co-optimisation that derives the line from the dispatch itself.
CM revenue by duration de-rating factor; unknown auction years modelled as a bounded random walk, disclosed in the manifest.
A per-TNUoS-zone × duration proxy, measured from the CE GB BESS Index — not dispatch-based in v1, and labelled as such everywhere it appears.
Plain-English everywhere else; this section for anyone vetting the maths. The full methodology, the index and the backtest evidence are live in-app.
Iterate the price-bias map to convergence under a KM harmonic step schedule — plain Picard provably fails beyond ~2 GW fleets. A residual monitor switches to a μ-regularised dispatch (a contraction) if monotone decrease stalls. Non-converged days are flagged, excluded from headline aggregates, and recorded — never quietly averaged in.
A fast parametric GB supply curve for interactive runs, or a scenario-grade whole-GB half-hourly merit-order LP — generator classes, curtailable renewables, interconnectors, unserved energy at VOLL £10,000/MWh — with the price taken as the dual of the power balance.
The price level is pinned to a chosen basis: the WP01 spec curve, an equilibrium-refit to realised 2024 Elexon prices, or a trailing-12-month anchor that recalibrates to the latest settled market and flags itself STALE past 35 days. Two anchors are two different price bases — the app refuses to compare them silently.
Finance runs on the multi-year cannibalisation-aware revenue trajectory — re-solving the equilibrium at sample years against the growing fleet — and reports the minimum DSCR with its binding year (typically late in the debt term). P50 target 1.40×, P90 lender covenant 1.10×, with a Monte-Carlo P90 on the same track shape.
CE-CENTRAL (18 GW BESS by 2030), CE-HIGH-RES (25), CE-LOW-GAS (14), CE-CONSTRAINED (12). Years 2025–2035, half-hourly across 26 representative weeks.
CE-CENTRAL on the DESNZ FFPA 2025 gas paths A/B/C — pinned, hashed, and carrying the DESNZ "not a forecast" caveat verbatim on every surface.
Set BESS GW by 2030, gas/carbon multipliers, demand growth, and paste per-year gas/carbon paths. Typed in, never fetched; disclosed as customer-supplied; carried in the run manifest so it stays reproducible.
CECadence is benchmarked against the CE GB BESS Index — our own index built from Elexon BMRS and NESO open data, no third-party benchmark licence. The 2022–25 backtest is committed and reproducible live in-app. 2025 is the pre-registered held-out validation year: it fails its tolerances, and the failure is published as-is. Cells that were fit to the targets they're graded against are badged fit (in-sample), never "pass".
Real Elexon backtest, committed and reproducible from the Methodology tab
2025 pre-registered as the validation year — badged, and failing, in the open
Every run keyed (scenario_id, code_version, timestamp) with a persisted manifest
What v1 does not do, stated plainly: the Balancing Mechanism line is a statistical zone proxy, not dispatch-based; price formation is single-zone GB (nodal via CEForesight is v2); demand/wind/solar profiles are synthetic-but-calibrated to the WP01 price-duration targets. The backtest bounds what these simplifications cost.
Every solve runs through one solver-abstraction layer at the HiGHS-compatible API surface, so the backend is a single config line — CE_SOLVER=meridian|highs|auto. CEMeridian, our native Rust solver, passes the LP-value-and-duals correctness gate against HiGHS on every push; auto resolves to HiGHS by default until the warm-start performance gate is signed off in production.
Appraise sites at the equilibrium, compare a portfolio head-to-head, and take a signed, reproducible bankable report into an investment committee.
The honest held-out disclosure is the feature: interrogate the assumptions, read the P90 DSCR covenant on the declining track, and replay any run from its manifest. On-prem and air-gapped for data sovereignty.
The cannibalisation sweep, sensitivities and Monte-Carlo bands for market views; the zone overlay for where revenue actually lands. Live as SaaS at cadence.compoundingenergy.com.
CECadence is in production at cadence.compoundingenergy.com — whole-GB scenarios, candidate-site appraisal, the bankable report, and the full validation dossier in-app. On-prem and air-gapped deployments available for data-sovereign teams. Talk to us about plans and access.
CECadence is built on our working-paper series. Both BESS papers are published in full — read them below.
The variational formulation, the convergence theorem behind the Krasnoselski–Mann iteration, and the bias result that anchors the product.
Extends the fixed point across the full stack with participation-bounded reserves, validated against the CE GB BESS Index.