Liver graft assessment is a critical and complex step in the transplantation process, traditionally reliant on the subjective judgment of experienced surgeons, based on initial donor data and macroscopic evaluation. In a previous study, we developed a machine learning-based expert system trained on clinical data to assist in the early decision-making process of liver graft suitability. Using donor information collected in the official Liver Donation Protocol (LDP), the model demonstrated promising predictive performance, particularly in identifying transplantable grafts that were otherwise discarded. This study aims to externally validate and expand the model using newly collected internal data and a large-scale external dataset provided by the Spanish National Transplant Organization (ONT). By evaluating the model’s generalization to these additional datasets and retraining it with combined data sources, we assess its robustness, adaptability, and potential for broader clinical application. Furthermore, we explore alternative modeling strategies to enhance predictive performance and reliability. This work represents a critical step toward refining and scaling an AI-driven tool to support transplant surgeons in early-stage graft evaluation using standardized donor variables.