Answers are grounded in this paper and cite their sources.
Auditable maintenance prioritization for constructed facilities must link structural evidence, probabilistic updating, resilience consequences, information quality, and resource-constrained action. This paper formulates a modular Bayesian risk–resilience decision score: a bounded, monotone, relabeling-invariant combination of calibrated limit-state failure probability, consequence, a standing recovery-time exposure, and an epistemic term, linked to value-of-information routing and constrained action selection. Observation-driven event probability is distinguished from physical failure probability. A simulation protocol generates the physical failure probability independently of the observational count model (300 assets, 50 seeds) and scores policies by calibration error and regret relative to a perfect-information greedy reference. Three results follow. Cross-fitted recalibration removes the gross miscalibration of the count-based probability, which remains only marginally more skillful than the base rate. Under a common action rule, no significant difference was detected between the reported configuration and a severity-based RCM-like rule at the reported budget; the advantage over probability–consequence, greedy, and TOPSIS baselines comes from the recovery term and cost-aware selection. A condition-informed failure-probability model calibrated against the physical benchmark, with the score reduced to its probability, consequence, and recovery terms, lowers regret by factors of 2.5–14 across simulated hazard regimes. All evidence is synthetic; field validation remains future work.
No takes yet. Share an insight, caveat, or question.
Rojas et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: