The roadmap is organized around making the research prototype more reproducible, realistic, and useful for energy-operation experiments.
- Keep the public benchmark reproducible.
- Improve documentation and onboarding.
- Make assumptions and limitations visible.
- Encourage focused contributions around tests, evaluation, and energy-domain examples.
- Add live or user-supplied tariff and carbon-intensity adapters.
- Add probabilistic forecasting with calibration and coverage monitoring.
- Add equipment-level constraints and operational feasibility checks.
- Add experiment artifacts and a small model-comparison report.
- Add a scheduled retraining example with drift thresholds.
- Explore mixed-integer scheduling for discrete equipment decisions.
- Add persistent storage for forecasts, schedules, and evaluation runs.
- Add production observability and deployment examples.
- Validate the workflow against an industrial or building-energy dataset.
- Forecasting must serve a decision.
- Every result needs an evaluation context.
- Constraints and assumptions are part of the model.
- A reproducible research prototype should be honest about production gaps.