Engineering simulation faces a fundamental tension: high-fidelity methods deliver accuracy but demand hours of computation and specialized expertise, while simplified approaches enable rapid assessment but sacrifice the precision required for consequential decisions. We propose decision-appropriate precision—a framework that matches computational investment to decision requirements rather than maximizing accuracy universally. The framework employs four discrete fidelity levels: (L1) pre-computed lookup tables achieving sub-millisecond response with ±15–30% accuracy for feasibility screening; (L2) physics-informed empirical models providing millisecond evaluation with ±8–15% accuracy for parametric exploration; (L3) machine learning surrogates delivering ±3–8% accuracy in tens of milliseconds for design optimization; and (L4) full-physics simulation ensuring ±1–5% accuracy for regulatory certification. Fidelity selection considers accuracy requirements, time budgets, user expertise, and decision stakes—routing queries to appropriate computational methods automatically. We demonstrate the framework through application to water-based cellular infrastructure, a novel domain presenting fluid-structure interaction, cellular network topology, and multi-scale organization challenges. The case study illustrates how unified workflow supports decisions from binary feasibility assessment through regulatory certification, enabling a single platform to serve users from casual exploration to professional engineering. The framework reframes accuracy-performance trade-offs as resource allocation rather than quality compromise. Lower fidelity for feasibility screening is not "dumbed down" simulation but decision-appropriate precision that reserves intensive computation for decisions genuinely requiring it. Recent advances in multi-fidelity neural operators and graph-based surrogates—demonstrating speedups exceeding 7,000× while maintaining physical consistency—validate this approach and provide the technical foundations for practical deployment.
James Otto Danenberg (Wed,) studied this question.