T-cell antigen receptor (TCR)-based therapies utilize TCRs that specifically target distinct peptide-Human Leukocyte Antigen complexes (pHLA). Successful binding of TCR to a given pHLA activates T cells to kill the target cell. Assessing TCR cross-reactivity to unintended targets is crucial to prevent clinical toxicities. Despite past failures resulting in severe toxicities, preclinical safety methodologies have improved through retrospective analyses. However, most studies focus on individual TCRs or small panels with the same pHLA target, lacking a comprehensive strategy for various TCRs. Our goal is to develop a robust in silico/in vitro model predicting and validating TCR cross-reactivity using a diverse TCR panel. This includes TCRs with past clinical toxicities, those proven safe yet effective, and innocuous TCRs without clinical experience. To capture the intrinsic cross-reactivity diversity, our panel includes naturally occurring and affinity-enhanced TCRs recognizing the same pHLA. In our predictive model, we deploy the X-scan technique to assess TCR-T cell activity following exposure to a panel of altered peptide ligands (APLs), allowing to evaluate the specificity of the TCR for the original pHLA and predict TCR degeneracy and cross-reactivity. Using primary human T cells engineered to express the TCR, we measure activation markers, cytotoxic activity, and effector molecule secretion. This data informs a computational model predicting cross-reactive peptides, which are experimentally validated within the human proteome. These refinements aim to establish a robust lab-in-the-loop model, guiding the development of a next-generation platform for derisking TCRs.