Adaptive vertical farms with adjustable shelf spacing require schedules that remain feasible as crop canopies expand and realized growth deviates from nominal assumptions, which can trigger facility-level vertical-capacity overloads after schedule release. We present a digital-twin-enabled controller that couples daily rolling-horizon optimization with a bounded feasibility-repair policy that restores feasibility through explicit, auditable schedule edits. Each day, a growth-aware time-indexed mixed-integer linear program (MILP) selects crop starts subject to shelf exclusivity, crop-mix bounds, and a facility-level vertical-capacity constraint derived from a triangular growth proxy. Upon release, the digital twin assigns each accepted batch a realized maximum height sampled at the decision boundary and stores it as a batch attribute, ensuring consistent capacity accounting over time. During execution, Multi Shift Feasibility Repair (MSFR) monitors predicted daily height usage and, when overloads occur, defers movable starts to the earliest conflict-free day on the same shelf within a bounded 30-day window using explicit shelf-conflict checks and prioritization by overload relief. This stabilizes execution without repeated full re-optimization, addressing schedule nervousness in rolling-horizon settings. The approach is evaluated in simulation for an eight-shelf facility over a 365-day horizon with two crops and batch-specific realized canopy maxima, using five stress scenarios with five Monte Carlo replicates per scenario. Active-day utilization ranges from 0.65 to 0.74 across scenarios. Residual overload energy is eliminated in four scenarios; under high noise, mean residual overload energy is 0.94 cm (95% confidence-interval half width 2.60 cm) when no feasible same-shelf deferral exists within the delay limit. The results quantify the controller’s ability to preserve feasibility under growth variability and operational stress while constraining the magnitude of schedule changes.