41-Day Live Challenge Tests EU-AI Act-Compliant Load Forecasting for German Grid

A 41-day live challenge evaluated a complete short-term load forecasting (STLF) pipeline for the aggregated German transmission-grid load, with results published on arXiv (2608.05018). The pipeline, built on the open-source Python library spotforecast2-safe, implements EU-AI Act requirements for safety-critical environments by design. It predicts 24 hourly load values for a target day using data from the European Network of Transmission System Operators for Electricity (ENTSO-E). The system includes anomaly detection, gap-aware data preparation, calendar and weather covariates, and a recursive multi-step forecasting approach. The study emphasizes that STLF is no longer solely an accuracy problem but also a software-engineering and compliance problem, as determinism, reproducibility, and auditability are engineering requirements for infrastructure designated as critical by European and German law. The challenge results highlight the practical application of AI regulations in energy forecasting.