Trust modeling is critical for the safe deployment of autonomous systems, yet existing approaches that rely primarily on historical performance data fail to capture dynamic operational contexts and real-time agent capabilities. This paper introduces a formal framework for modeling actual trust in Autonomous Delivery Vehicles (ADVs)—a context-aware trust model that evaluates an agent’s current ability, knowledge state, and commitment to task completion rather than relying solely on past behavior. We present a systematic refinement-based approach using Event-B formal methods to model trust in ADV task delegation scenarios. Our methodology progresses through five refinement levels, transitioning from an untrusted baseline model to a comprehensive trust framework that integrates three key dimensions: (1) strategic trust (capability verification), (2) epistemic trust (knowledge-based assessment), and (3) commitment trust (availability and willingness evaluation). Each refinement level addresses specific failure modes identified in traditional delegation systems where tasks may be assigned to incapable, unknown, or unavailable vehicles. The formal model is verified using the Rodin theorem prover with 93 proof obligations, achieving 90% automatic verification. Our approach demonstrates how actual trust can be systematically integrated into autonomous systems through correctness-by-construction refinement, ensuring that task assignments occur only when trust conditions are formally verified. The framework provides a foundation for trustworthy task delegation in multi-agent autonomous systems and offers insights for developing reliable AI-driven delivery networks.