Disrupted humanitarian supply chains require decisions that are fast, interpretable, and robust to operational deterioration. This paper develops a policy stress-testing framework for relief-network planning under disruption by transforming multi-objective optimisation outputs into policy classes that can be compared under different operating scales. A scenario-based humanitarian supply chain model is solved using a preference-guided extension of NSGA-II, and the resulting non-dominated solutions are translated into lean, balanced, and resilient policy classes. Three test cases are considered, representing local, regional, and corridor-level relief operations. The numerical study shows that preference-guided search consistently identifies policies with better service continuity and lower resilience loss than classical baselines, while the graphical stress-testing analysis clarifies the trade-off between additional cost and service protection. The study demonstrates how disruption data can be converted into policy-relevant insight for humanitarian logistics planning.