This paper investigates the challenge of balancing performance and transparency in time-series models for battery health monitoring. In this domain, AI models are essential for predicting metrics such as Remaining Useful Life (RUL) and State of Health (SOH). However, achieving high performance while maintaining interpretability remains a critical challenge, particularly in safety-critical applications. Advanced time-series models like Temporal Fusion Transformers (TFTs) provide high predictive accuracy but are often complex and less interpretable. We demonstrate how incorporating techniques such as attention score visualization and SHAP (Shapley Additive Explanations) can enhance their transparency, offering insights into key feature contributions. Conversely, Decision Trees, while naturally interpretable, may lack the accuracy of deep learning models. However, they can be improved through methods like time-windows and Gradient Boosting to better capture temporal patterns without losing their explainability. The paper also examines potential models such as Cellular Neural Networks (CeNNs), which excel in real-time processing and can be optimized with global attention mechanisms to balance local and global interpretability. We outline essential requirements for developing explainable battery health monitoring models, presenting a set of best practices for selecting suitable models and explainability techniques that align with FAIR (Findable, Accessible, Interoperable, Reusable) and XAIR (Explainable-AI-Ready) principles while addressing this trade-off. These recommendations are based on reviewing existing studies that have enhanced model interpretability or accuracy to address the performance-transparency trade-off.