Trust and Reputation - related but not the same

I'm sure there's awareness of this but I've not seen it explicitly written down anywhere...

Trust and reputation is distinct but closely coupled. A lot of computational trust/reputation models include aspects of both but I've yet to see one which tries to separate the two.

Trust is implicit in actions taken by an agent. The management of trust is thus the management of policies regarding specific actions to take given the parameters that are important to that situation. For example, to buy a DVD player on eBay implies that certain criteria governing the policy to buy from the seller has been met - i.e. the buyer trusts the seller enough to buy what he's selling.

Reputation is information that is used as one of the parameters in deciding whether to trust. In other words, trust management takes reputational information as one of it's inputs and then decides whether an action should be executed (based on various other situation specific inputs, such as minimum trust level, risk, motivation, etc.).

For any application with automated trust reasoning, there are at least two trust management components and one reputation evaluation component at work.

Trust manager 1 belong to the application. It makes decisions for the application on whether certain actions should take place. For example, it decides whether one should buy the DVD player from a given eBay seller. As it uses reputational information as one of its inputs, it queries the reputation evaluator.

The reputation evaluator collects reputational information about the seller from other agents and makes some assertion about the reputation of the seller in question. However, to do this, it must also reason about the trustworthiness of the recommenders. This is the job of Trust Manager 2 - it contains policies that govern the dynamics of trust relationships with other recommenders and how their recommendations should be consumed.