Code and results: When Do Time Series Foundation Models Help Forecast Wind Energy Droughts? A Preregistered Evaluation

Code, NWP covariates, per-window scores and logs for the article "When Do Time Series Foundation Models Help Forecast Wind Energy Droughts? A Preregistered Evaluation" (Q. Dai, submitted to IEEE Transactions on Knowledge and Data Engineering). The study evaluates time series foundation models (Chronos-2, TimesFM 2.5, Moirai-2) against climatology, model output statistics, quantile regression and a recurrent network for probabilistic forecasts of the cumulative wind energy deficit in 16 regions in Europe, Australia and the US. The protocol was registered on OSF (https://osf.io/ph7nr/) before the prespecified main analysis of the sealed test period (January 2024 to September 2026) was scored once. Subsequent post-hoc analyses included additional forecasts and rescoring. The archive contains the code of every step, NWP covariates derived from ECMWF IFS HRES forecasts (CC BY 4.0), the per-window scores of all methods, run logs, and the post-hoc fairness analyses; the registered protocol is on OSF. Observed wind power is not redistributed; the released scripts download it from its sources. See README.md.