Consensus‑Aware Bridge Maintenance Planning with Auditable Evidence and Multi‑Stakeholder AI Evaluation

Bridge-maintenance planning under annual budgets must balance safety, cost, disruption, and equity while remaining explainable in public review. Working with engineers from a railway operator and a major general contractor, we translate three adoption barriers, namely hidden burdens, justification overhead, and value conflicts, into measurable criteria. We propose a decision-ready framework that combines a lightweight digital twin, constrained multi-objective optimization, synthetic Virtual Citizens for distribution-sensitive burden metrics, and an evidence-restricted LLM evaluator for multi-persona acceptability and disagreement. The LLM scores only auditable plan-summary JSON with deterministic decoding and caching. In an offline case study on public data, the framework outputs reviewable plan artifacts that support deliberation, audit, and accountability rather than replacing human decision makers.