Open pit blast design must deliver safe, compliant outcomes (eg controlled airblast) while meeting production targets, such as desired fragmentation for loading, crushing, and downstream throughput. In practice, blast response is nonlinear and site dependent because design parameters (burden, spacing, stemming, charge distribution and timing) interact with rock mass behaviour and structural conditions. This motivates the use of machine learning (ML) approaches for prediction; however, operational adoption is often limited when models behave as black boxes and cannot justify why a prediction changes under design modifications. This work develops interpretable extreme gradient boosting (XGBoost) regression models for two key blast outcomes: airblast level (dBL) and fragmentation median size (X50, mm). Explainability is provided using SHapley Additive exPLanations (SHAP), enabling both global identification of dominant drivers and local diagnosis of individual blast predictions (SHAP is an explainable AI technique that uses Shapley values to quantify how each feature contributes to a prediction relative to a baseline, enabling interpretation of both overall model behaviour and individual predictions).