Position Paper: Post-Solve Robustness in Decision Engines: Feasible Regions and Smoothness Under Perturbations

Researchers propose a post-solve robustness framework for Mixed-Integer Linear Programming decision engines, addressing the gap between theoretical optimal solutions and real-world deployment where parameter perturbations can invalidate feasibility. The work calls for standardized auditing of solved problems to measure how solutions perform under small cost, demand, and resource variations.

This position paper tackles a fundamental but overlooked problem in optimization: the brittleness of nominally optimal solutions in practice. MILP solvers generate mathematically optimal plans for critical industrial systems—supply chains, energy grids, manufacturing schedules—yet these solutions often fail when real-world conditions diverge from solve-time assumptions. A 5% shift in input costs or a minor demand fluctuation can render a solution infeasible or trigger a discontinuous jump to an entirely different plan. The authors reframe this as a missing evaluation layer rather than a failure of robust optimization or stochastic programming. Their contribution formalizes two diagnostic objects: epsilon-near-optimal feasible neighborhoods that measure parameter tolerance around an incumbent solution, and solution smoothness that captures whether nearby combinatorial alternatives remain competitive. The framework synthesizes insights from sensitivity analysis, neighborhood search, adversarial testing, and learning-based verification, positioning post-solve robustness as a solver-backed certification step. For industrial operations and AI-driven decision systems, this matters significantly. Decision engines deployed in high-stakes environments—grid operations, portfolio optimization, resource allocation—require transparency about solution fragility. The proposed reporting template and evaluation protocol would make robustness a measurable, first-class output alongside optimality. This directly impacts trust in learning-enabled systems and the viability of end-to-end optimization pipelines in production. The work signals growing recognition that mathematical optimality alone is insufficient; systems must demonstrate stability under realistic perturbations.

• →Post-solve robustness audits bridge the gap between theoretical optimality and real-world solution stability under parameter perturbations.
• →The framework formalizes epsilon-near-optimal feasible neighborhoods and solution smoothness as core diagnostic objects for decision engine transparency.
• →This layer complements rather than replaces existing robust optimization and stochastic programming approaches.
• →Standardized robustness reporting and evaluation protocols would enable comparison and trustworthiness assessment of AI-driven decision systems.
• →Industrial optimization pipelines need solver-backed certification of solution fragility to manage high-stakes deployment risks.