In this paper we develop a novel probabilistic model of computational trust that allows agents to exchange and combine reputation reports over heterogeneous, correlated multi-dimensional contracts. We consider the specific case of an agent attempting to procure a bundle of services that are subject to correlated quality of service failures (e.g. due to use of shared resources or infrastructure), and where the direct experience of other agents within the system consists of contracts over different combinations of these services. To this end, we present a formalism based on the Kalman filter that represents trust as a vector estimate of the probability that each service will be successfully delivered, and a covariance matrix that describes the uncertainty and correlations between these probabilities. We describe how the agents’ direct experiences of contract outcomes can be represented and combined within this formalism, and we empirically demonstrate that our formalism provides significantly better trustworthiness estimates than the alternative of using separate single-dimensional trust models for each separate service (where information regarding the correlations between each estimate is lost).