Abstract
<title>Abstract</title> <p> Exact solvers, metaheuristics, and priority dispatching rules are routinely compared on the deterministic job-shop problem, where an exact solver is, by definition, the quality ceiling. Real shop floors, however, are subject to machine breakdowns, stochastic processing times, and the dynamic arrival of new jobs, and it is far from obvious that a schedule which is optimal under nominal data remains preferable once it must be executed under disruption. This study quantifies that question directly. Using a discrete-event, predictive–reactive evaluation harness, we execute the committed schedules produced by an exact constraint-programming solver (CP-SAT) and by a genetic algorithm (GA) under a right-shift repair policy, alongside nine online priority dispatching rules, across nine disruption regimes (three severities each of breakdowns, stochastic times, and arrivals), on sixteen Fisher–Thompson and Lawrence benchmark instances with thirty replicate scenarios per cell. Robustness is measured by a relative-degradation index and, decisively, by realized solution quality against the certified optimum. The central finding is an inversion: the exact-optimal schedule, best under nominal data (mean 3.1% above optimum versus 12.1% for the best rule), is the <italic>least</italic> robust method tested (Friedman rank 10.1 of 11, versus 3.9–5.8 for the rules; all pairwise Wilcoxon <italic>p</italic> < 10⁻⁴), because an optimally compact schedule has no slack to absorb disruption. Its nominal advantage does not survive execution: under machine breakdowns the most-work-remaining and most-operations-remaining rules overtake it in realized quality, and under dynamic arrivals the online rules dominate it outright. We conclude that exact optimality and disruption robustness are conflicting objectives on the dynamic job shop, and that for disruption-prone environments a simple, computation-free dispatching rule is not merely a fallback but frequently the superior deployed policy. </p>