Wind farm power control helps the grid maintain balance and avoid blackouts. In Europe this is often achieved through implicit balancing, where a wind farm adjusts its power set‑point using information published by the Transmission System Operator, anticipating the imbalance price. Thus the farm assists system balance restoration while reducing its own imbalance costs caused by forecast errors.\ Implicit balancing is challenging because the settlement price is unknown beforehand and can be influenced by the chosen control strategy. Reinforcement learning (RL) methods could address this dynamic decision problem, but their lack of interpretability hinders deployment on real wind farms.\ We therefore propose a policy distillation framework based on Explainable Boosting Machines (EBM), a variant of generalized additive models (GAM) that yields transparent control policies. To increase the value captured by the EBM, we introduce a value‑oriented perspective to policy distillation, inspired by energy‑forecasting research: the distillation objective maximizes expected returns rather than merely mimicking actions.\ We also incorporate expert knowledge during model design, e.g., by constraining functional forms or selecting informative features, so that the resulting strategy remains justifiable to stakeholders.\ Experiments on real data from the Belgian grid and an offshore wind farm show that the value‑oriented, expert‑augmented controller reduces imbalance costs by 82%, outperforming a price‑threshold controller and a decision‑tree‑based distillation baseline.\ Review: This work cleverly merges interpretable models with value‑oriented distillation, delivering a deployable and transparent wind farm control solution for real‑world power systems, and offers a promising pathway for applying RL in energy domains.