The long-term performance of montmorillonite-based engineered barriers in nuclear waste repositories and contaminated groundwater systems depends critically on understanding how sequential heavy-metal sorption alters interlayer chemistry, hydration equilibria, and macroscopic transport properties. Despite decades of research on single-cation systems, the coupled physicochemical processes governing multi-step cation exchange—where incoming divalent species (Co²⁺, Ni²⁺, Mg²⁺, Cu²⁺, Zn²⁺, Cd²⁺, Pb²⁺, Ba²⁺) progressively displace weakly-bound Na⁺ from both external surfaces and interlayer galleries—remain poorly constrained, particularly under conditions that generate coexisting hydration states. This gap is addressed by integrating high-resolution XRD 00l profile modeling of Wyoming-type Na-montmorillonite (SWy-2) across systematic exchange sequences with interpretable machine-learning architectures trained on 600 physics-enriched synthetic records combining crystallographic observables (d₀₀₁ spacing, 0W/1W/2W layer-type abundances, Rietveld Rₘₚ values, coherent domain lengths) and thermochemical descriptors (hydration enthalpy, ionic potential, polarizability). XRD refinements reveal that sequential exchange does not converge toward homogeneous homoionic but instead stabilizes dynamically disordered mixed-layer structures with 15–35% residual Na⁺ occupancy driving non-monotonic d₀₀₁ trajectories that follow Boltzmann-type sigmoid curves rather than linear replacement trends, with effective CEC losses reaching 40% for large, weakly-hydrated cations (Pb²⁺, Ba²⁺) due to incomplete interlayer accessibility. Ensemble machine-learning models (Random Forest, Gradient Boosting, Neural Networks) trained on this augmented feature space achieve robust cross-validated performance (R² = 0.90–0.92 for fractional uptake; 84–86% accuracy for competitive selectivity ranking), with SHAP decomposition demonstrating that hydration free energy and ionic radius dominate early-stage exchange kinetics while XRD-derived structural disorder metrics become critical predictors of late-stage degradation modes including irreversible 2W→1W dehydration, basal-spacing collapse below 12.5 Å, and the emergence of nano-scale permeability pathways. By embedding crystallographic disorder signatures into predictive models, our approach transcends empirical correlation and enables mechanistic forecasting of barrier aging under realistic poly-ionic exposure scenarios, providing quantitative risk assessment tools for repository safety analysis and adaptive monitoring strategies in deep geological disposal environments over millennial timescales.