Short, reproducible technical notes on U.S. load-forecasting and data-quality problems. One question, one chart, the code behind it. Each note is published first at cardinalgrid.com/notes, archived on Zenodo with a DOI the same day, and republished on Medium, LinkedIn and Energy Central with the site as the canonical version.
| # | Note | Target |
|---|---|---|
| 1 | How well do U.S. balancing authorities forecast their own load? A 2015–2026 scorecard from EIA-930 | October 2026 |
| 2 | Winter Storm Elliott, revisited with public data | October 2026 |
| 3 | Were the "bad data" flags in the January 2024 Arctic storms really bad data? | November 2026 |
| 4 | Step loads: detecting data-center-scale structural breaks in ERCOT and PJM zonal load | November 2026 |
| 5 | From forecast error to reserve cost: a transparent back-of-envelope for ratepayers | December 2026 |
| 6 | How much of EIA-930 is imputed, and where? | December 2026 |
| 7 | Cold-snap regimes: why one model per season fails, and a probabilistic alternative | January 2027 |
| 8 | State of U.S. load forecasting 2026 | January 2027 |
Titles may change as the analysis is done; dates are target months.
notes/
01-ba-forecast-scorecard-2015-2026/
README.md the note
notebook.ipynb reproducible analysis
data/ pinned snapshot or download script
02-.../
Each folder has a requirements.txt and a notebook that downloads (or pins) the public data it uses. Run it top to bottom.
Corrections are made here and on the site first, then noted in the changelog of the archived Zenodo record.
Text and figures: CC BY 4.0. Code: Apache-2.0.