version string | source string | source_index_url string | timezone string | timestamp_convention string | load_definition string | scope string | cases list |
|---|---|---|---|---|---|---|---|
ercot-native-load-cases-v1 | ERCOT Hourly Load Data Archives | https://www.ercot.com/gridinfo/load/load_hist | America/Chicago | Interval start, converted from ERCOT market-local Hour Ending labels; each row spans one hour. | ERCOT published hourly native load. This series differs from the forecast-zone load data used by the probabilistic research candidate. | Observed historical load with user-defined counterfactual assumptions. No live data or probabilistic forecast is produced. | [
{
"id": "summer-2023",
"title": "Summer load · August 2023",
"note": "A high-load summer window. Compare a fixed capacity assumption with a proportional load reduction.",
"default_capacity_mw": 80000,
"start_date": "2023-08-09T00:00:00",
"end_date_exclusive": "2023-08-12T00:00:00",
"sour... |
GERT Historical Lab — ERCOT load observations
A deterministic teaching exercise: reproduce a grid-risk calculation from 216 hours of public ERCOT native-load observations (three 72-hour windows: Aug 9–11 2023, Apr 10–12 2023, Feb 14–16 2021).
The learner sets two assumptions — available capacity (MW) and uniform load reduction (%) — and the lab computes hourly gap power and total gap energy. It is not a validated forecast and not a representative validation sample: its purpose is to make the assumptions behind a grid-risk number visible, and show how much they move the answer.
Honest status (2026-09-30): the probabilistic candidate ercot-delta-quantile-v1.4 was publicly
rejected for production after missing its predeclared 0.03 calibration tolerance at P90/P95/P99
(see docs/DECISION_RECORD_ERCOT_v1_4.md). The Historical Lab is an untested pilot teaching exercise.
Files
| File | What |
|---|---|
data/historical-load.csv |
216 hourly rows: case_id, interval_start, recorded_load_mw, source_archive_sha256 |
data/historical-cases.json |
Case metadata, sources, methodology parameters |
data/historical-cases.sha256 |
SHA-256 provenance for the JSON |
docs/HISTORICAL_LAB.md |
Methodology: data sources, provenance, rebuild instructions, math, interpretation limits |
docs/DECISION_RECORD_ERCOT_v1_4.md |
Public record of the rejected v1.4 probabilistic candidate |
scripts/verify_historical_report.py |
Independent checker: recomputes every hourly value from a JSON report (Python 3.10+, stdlib only) |
Reference calculation (hand-checkable)
Summer-2023 case, capacity 80,000 MW:
- 0% reduction → 21 hours above capacity, gap energy 66,683 MWh
- 5% reduction → 7 hours above capacity, gap energy 4,458 MWh
Per hour, with recorded load L, capacity C, reduction r:
adjusted = L * (1 - r / 100)
gap_original = max(0, L - C)
gap_adjusted = max(0, adjusted - C)
Provenance
Source: public ERCOT native-load archives, repackaged with SHA-256 checksums. Also archived with DOI 10.5281/zenodo.22975251 (MIT license). Try the interactive lab: https://gert-d.vercel.app/history
Citation
Goehner, Dresden E. (2026). GERT Historical Lab dataset (ERCOT teaching exercise).
Zenodo. https://doi.org/10.5281/zenodo.22975251
- Downloads last month
- 89