CoreSense is an architecture for robot decision-making that includes a conflict-aware belief gate to ensure auditable and safe decisions. It evaluates evidence and permits or escalates decisions. Evaluation shows reduced unsafe proceeds and improved performance in various experiments.