When is electricity-demand inference warranted? A regime-conditioned, feasibility-first analysis of national grid demand

Background National electricity demand is increasingly shaped by calendar structure, renewable generation, interconnector flows and demand-side flexibility. Long historical records therefore do not automatically establish that every current demand state is comparable with earlier observations. Methods The analysis included 278,512 UK settlement-period observations from 2009 to 2024. A regime-conditioned admissible-state operator retrieved only strictly prior comparable observations defined by clock regime, holiday status, season, wind regime and interconnector regime. Inference was released when prior support met nmin=25; otherwise, the operator returned HOLD. Released observations were evaluated against a prior 5th-95th percentile demand envelope. Sensitivity analyses varied support thresholds, envelope width and regime grammar. Baseline comparisons used pooled envelopes, clock-season envelopes, anomaly rules, seasonal naive, seasonal autoregressive and Random Forest models over 2022–2024. Results The prior-only primary specification identified 319 state cells, of which 270 eventually reached release support. Row-level HOLD was 2.63 %. Among 271,177 released rows, breach rate was 16.76 % (95 % CI 16.62–16.90), above the nominal 10 % expectation, indicating nonstationary regime-sensitive departures. Baselines produced point forecasts for all evaluable rows but did not identify inferential non-release states. Conclusion Feasibility-first electricity-demand analysis is not a forecasting replacement. It is a release-readiness layer that distinguishes supported inference, breach and HOLD under active operational regimes.