Day-Ahead Load Forecast Accuracy of U.S. Balancing Authorities, 2015–2026: A Scorecard from Public EIA-930 Data (Cardinal Grid Notes, No. 1)

Balancing authorities in the contiguous United States report to the U.S. Energy Information Administration, through Form EIA-930, both the hourly demand that occurred and the day-ahead demand forecast they produced the previous day. This note scores those forecasts for the full public record, from 2015-07-01 to 2026-09-10, with explicit data-quality rules. Of 72 BAs in the files, 58 report both series and 42 are eligible for ranking. The demand-weighted day-ahead MAPE across eligible BAs is 3.56%; at the hour of the daily peak the forecast was below actual demand on 59% of eligible BA-days, a share that has increased since 2019. The highest error values are dominated by reporting problems rather than forecasting problems. Code, results and the event calendar are open. Web version: cardinalgrid.com/notes. Source and reproducible analysis: github.com/cardinalgrid/notes. Version 1.1 (2026-09-14). Corrected two source attributions in the introduction: the review of the January 2025 Arctic events (FERC/NERC, April 2025) replaces a statement attributed to the January 2024 review that the source does not contain, and the NERC 2025 Long-Term Reliability Assessment is now quoted as written. Numbers, figures and conclusions are unchanged. Version 1.2 (2026-09-14). Same content as version 1.1 (corrected source attributions in the introduction; numbers, figures and conclusions unchanged); this version prints its own DOI in the header and in the web version.