The mathematics behind the products, free to read. Every result can be reproduced from a stated scenario, code version and timestamp.
WP 2026-01Cannibalisation as a Fixed Point. Why naive battery-revenue forecasts overstate by 150–440% at 4–10 GW fleets, and the joint fleet–price–dispatch equilibrium that corrects them, solved by Krasnosel’skii–Mann iteration. The basis of CECadence.
WP 2026-02The GB BESS Revenue Stack. Wholesale energy and the six NESO dynamic-response products co-optimised under one shared power and state-of-charge budget. It shows why the naive stacked sum can’t be realised, with a fully reproducible stylised GB case study.
WP 2026-03Stabilised Benders with N-1 Cuts. Benders masters stabilised with a level bundle and Magnanti–Wong cuts, with embedded N-1 contingency cuts, for security-constrained nodal expansion. One of the foundations of CENovaSage's capacity-expansion engine.
WP 2026-05The Cannibalisation of Flexibility. Data-centre demand response priced at the same equilibrium that repriced batteries. Price-taking valuations overstate GB flexibility value by 1.15–1.73× between 2 and 10 GW, the shallow ancillary stack saturates first, and 10 GW of flexible load takes 21% off the battery fleet's wholesale-plus-response revenue.
WP 2026-06Load Growth and the Un-cannibalisation Ledger. The other side of the data-centre question. Load that the system is built to accommodate lifts renewables’ capture (solar capture ratio 0.36 → 0.58 at 10 GW, worth £1.7bn a year) and fills the troughs the battery fleet charges in (−26%, or −38% with half the load flexible). It helps one fleet and competes with the other at every penetration tested.
WP 2026-07Who Cannibalises Whom: Duration Classes at the Storage Equilibrium. One-, two- and four-hour batteries solved jointly across 63 fleet mixes. Every entry of the cross-cannibalisation matrix is negative, own-effects are steepest for four-hour, and which class loses most depends on whether you count pounds or share.
WP 2026-08From Bias to Bankability: Cannibalisation, DSCR Covenants and the Term Structure of Storage Revenue. A 2027 two-hour battery followed through a fifteen-year tenor at equilibrium. The gap to today's price shape opens in the first three years and stays; debt sized on today's prices breaches a 1.10× covenant in year three; debt sized at equilibrium is 25% smaller (60% on a fast build-out).
NOTE 2026-01Training Equals Production. A short note on calibration discipline: models are retrained on the same artifacts they serve, which keeps calibration drift bounded by what users actually saw.
NOTE 2026-02The Graded Record. The pre-registered CECadence backtest record in full, 2022–2026H1. Rules were frozen in dated commits before the data was fetched; three of five annual-mean gates were met; both held-out windows failed on the scarcity tail and are published as they landed; five re-shape attempts were rejected under their own rules.
NOTE 2026-03The Graded Scorecard. CompoundVision's three public accuracy scorecards, frozen on stated dates and reported in full. On the 6 September scorecard, site calibration against Elexon metered output cut out-of-sample error for 64 of 65 scored GB wind farms (median 42%, capacity-weighted 45%), and solar beat persistence in 29 of 30 regions and wind in 24 of 32. Every region and every exception is printed.
NOTE 2026-04Where the Balancing Money Goes. GB Balancing Mechanism payments in 2025/26 traced to technology, region and lead party from Elexon settlement data, with NESO's cost categories and system flags. £1.76bn of £2.47bn went to gas plants in England and Wales paid to turn up; Scottish wind was paid £387m to turn down. Every unit is in the open data pack.
Technical due diligence: validation dossiers, benchmark provenance and methodology documents for every product are available under NDA where they aren't already public. Ask us.