The basis of a trust value is always the same: two or more parties commit to a contract (Ramchurn, Jennings, Sierra & Godo 2004) that defines an interaction (possibly composed of several distinct steps) as well as its expected result. The actual result of the interaction can be compared to what was expected according to the contract, thus yielding an experience for each party. Experiences can in turn be used to generate trust values which inform future interactions (see figure). The derivation of trust values is determined by a trust model. An additional interaction can be necessary to observe compliance with the contract.

In many circumstances, trust is measured as a numerical value, often ranging between 0 and 1. The trust model is basically an algorithm whose input, the experiences, are transformed into this value. Trust values should always carry a form of semantics. Such semantics can, e.g., be introduced by a clearly defined way the agents interpret trust values. In many cases, numerical values also provide an order that can be used to sort agents according to their trustworthiness. In such a case, it might not even be necessary to supply clear semantics for a trust value of 0.5 as long as it is clear that such an agent is worse than one with a trust value of 0.7.