Data centres and digital research infrastructures (RIs) increasingly need predictive capabilities to operate efficiently and reduce their carbon footprint. In GreenDIGIT, we build a MetricsDB-based pipeline and an ML prediction service to forecast site-level operational signals for several use-cases: brokering services, prediction systems, capacity planning and prediction. This challenge is a forecasting task based on a mature DIRAC-operated compute site (real identity withheld; released under a pseudonymous site ID). Participants predict near-future energy consumption and carbon footprint from historical time series. The task is realistic: it includes (i) daily/weekly seasonality, (ii) bursts/outliers, and (iii) irregular sampling and downtime gaps that require robust time-series handling. Impact: better forecasting enables proactive resource allocation, improved service availability, and measurable reductions in energy and carbon footprint through smarter operational planning.
