Accurate electricity price forecasts are critical for optimizing flexible consumption, such as electric vehicle (EV) charging. Existing work has shown that multi-day ahead forecasts can reduce consumer costs, but two practical challenges remain. First, exogenous variables such as renewable generation, load, imports, and fuel prices are often unavailable beyond day-ahead horizons, leading researchers to rely on error-based proxies. Second, many studies train models on fixed datasets, which may limit their ability to adapt to changing market conditions. This paper proposes two extensions to address these issues. We outline an approach to incorporate multi- day forecasts of exogenous variables directly into electricity price models, and we evaluate rolling and expanding window retraining schemes to account for non-stationarity and seasonality. Using Dutch market data, we assess how these design choices affect both forecast accuracy and the downstream performance of forecast-based EV charging optimization. The results highlight the importance of realistic data pipelines and adaptive training strategies for improving the reliability and relevance of consumer-oriented forecasting applications.