Reliable probabilistic electricity price forecasting constitutes an essential prerequisite for optimal battery scheduling in deregulated energy markets. This study presents a comparative evaluation of probabilistic forecasting methods for the Dutch EPEX day-ahead market, examining Quantile Regression (QR), LASSO-regularized Quantile Regression with Bayesian Information Criterion selection (LQR-BIC), and Natural Gradient Boosting (NGBoost) against a naive benchmark. Our experimental framework employs an evaluation design spanning six distinct forecast origins from 2018 to 2024, encompassing stable conditions, the COVID-19 pandemic, and the 2022 energy crisis. Results reveal that linear quantile methods achieve superior average probabilistic accuracy, while NGBoost demonstrates the best calibration in half of the evaluated periods once adapted to prevailing market conditions. However, NGBoost exhibits vulnerability during rapid regime transitions, with degraded performance at the September 2021 crisis onset. These findings suggest context-dependent model selection for battery scheduling, where QR offers robustness across conditions while NGBoost provides superior uncertainty quantification during stable or established regimes.