Conservation decisions under deep uncertainty require models that produce plausible ecosystem projections at a tractable computational cost. Sensitivity-guided model development and ensemble calibration are established in other fields, but remain uncommon in marine ecological projections. A workflow that uses distribution-based sensitivity analysis of Great Barrier Reef coral demographic data to guide model structure is presented here. Colony size was the dominant factor associated with growth and among the leading factors associated with survival; functional group and location-specific conditions also contributed. These findings informed Kora, a size-structured population model that tracks five coral functional groups using growth and background-survival relationships fitted to observations. Kora also represents heat-stress mortality, thermal-tolerance inheritance, reproduction, and recruitment. Ensemble calibration at two reefs reproduced observed coral-cover trajectories despite contrasting monitoring histories. The workflow provides a practical basis for uncertainty-aware projections where monitoring is sparse and rapid exploration of management scenarios is needed.