As part of the continuing development of the boiling water reactor in-core fuel management optimization code FORMOSA-B, the cold shutdown margin (SDM) constraint evaluator has been improved. The SDM evaluator in FORMOSA-B had been a first-order accurate Rayleigh quotient variational technique. It was deemed unreliable for difficult perturbed loading patterns (LPs) and thus was replaced by a high-fidelity, robust, computationally efficient evaluator. The new model is based on the solution of the one-group diffusion equation using approximate albedo boundary conditions for a three-dimensional, variable axial node, 10 × 10 assembly subregion around the stuck rod location. The fidelity and robustness of the model are first demonstrated by performing calculations on difficult perturbed LPs and for different plant cores. It is shown that the SDM reactivity is estimated within 40 pcm for the highest worth rod and that the speedup factors are 50 to 100 for small cores (and even more for larger cores) in comparison to the full-core three-dimensional simulations. Next, the successful implementation of the model in imposing the SDM constraint for FORMOSA-B's adaptive simulated annealing (SA)-based optimization strategy is presented. The results demonstrate SA's ability to remove large SDM violations (>700 pcm) along with thermal margin and critical flow constraint violations. Finally, the importance of having the SDM constraint on during optimization is shown by comparing results with a simulation in which the constraint is off.