| --- |
| license: mit |
| task_categories: |
| - reinforcement-learning |
| tags: |
| - electricity-markets |
| - battery-dispatch |
| - ercot |
| - benchmark |
| size_categories: |
| - n<1K |
| --- |
| |
| # ERCOT-RTCB v1 |
|
|
| A daily proxy benchmark for feasibility-aware battery dispatch |
| under ERCOT Real-Time Co-optimization Plus Batteries (RTC+B). |
|
|
| ## Dataset Description |
|
|
| 304-day daily panel (June 2025 to March 2026) with real-time LMP, |
| MCPC for seven ancillary-service products, weather, and calendar |
| features. Observed five-minute MCPC covers 7 in-window days; the |
| remaining 297 use a frozen folded-normal generator calibrated from |
| a 14-day public sample. Each row carries an mcpc_is_real flag. |
|
|
| ## Files |
|
|
| - sced_panel.parquet: 304-day daily panel |
| - leaderboard.json: baseline agent results |
| - sbr_results_500k.json: structural-break regret diagnostic |
| - croissant.json: MLCommons Croissant metadata |
| |
| ## Usage |
| |
| See the companion paper and code repository for the Gymnasium |
| environment, baseline agents, and reproduction instructions. |
| |
| ## License |
| |
| MIT |
