Limitations

What the model does not do — stated up front.

EVIDENCE · FIRST-CLASS

Every model has boundaries; these are stated in advance. Everything below is drawn from the engine’s own docs/05-validation.md and docs/08-risks-and-decisions.md.

Structural boundaries

  • No intra-GB network model. Transmission constraints are approximated by a single constraint-cost term (the B6 / Scottish-wind gap), not a power-flow solve. There is no nodal or zonal load-flow within GB (ADR-12).
  • No EMT-level stability. Stability is an aggregate swing-equation event simulation — inertia, RoCoF, frequency nadir — not an electromagnetic transient model.
  • Rule-based dispatch has no foresight. The default storage dispatch policy cannot see the future. This is mitigated, not hidden: results are also reported under a perfect-foresight LP policy, and the gap between the two is itself a reported finding.
  • Interconnector counterparties are aggregated. Continental counterparties collapse to three or four zones, not a full European model.

Convention-driven caveats

  • Embedded generation and solar circularity. Under the total-generation convention (D3), underlying demand is National Demand plus NESO’s half-hourly embedded-generation estimate, with embedded capacity modelled explicitly. A direct consequence: solar validation is partly circular on the NESO embedded estimate, because observed transmission-metered solar is zero. This is why solar is excluded from the headline validation table.

Data-scope boundaries

  • Public reproduction is GB-only. The redistributable data pack covers 12 of 14 scenarios. The two continental scenarios depend on ENTSO-E-derived data that may be used but not redistributed, so their reproduction is bring-your-own-token (see Reproducibility & data).
  • Weather record. Capacity factors and temperatures span 1985–2024. Years outside that window are not available.

What this is deliberately not

  • Not a network / power-flow model.
  • Not an electromagnetic transient simulator.
  • Not a live in-browser simulation (that is a deliberately decoupled later phase; today’s Run page is a stub).
  • Not a pricing-behaviour or subsidy-projection layer — those are separate, contested, and deliberately kept out of the instrument.
Note

If you find a boundary that is not listed here and that materially changes a published conclusion, that is a correction we want to make in public — see the Errata policy.