Chris Slater presented A Computational Model of Trust and Reputation today in class. The paper introduces three concepts--reputation, reciprocity, and trust--and how they relate to each other. We talk a lot about reputation and trust, but don't often consider reciprocity. They define reciprocity as a "mutual exchange of deeds (such as favor or revenge)."
In a reputation system focused on stopping blog comment spam, for example, the engine that calculates the score is calculating reputation, the threshold that you set in your software (e.g. moderate commenters with scores below 20) is the trust metric. Reciprocity is the probability that ham will be accepted and spam rejected.
An expectation of reciprocity and consistency in action is important if there's to be any social benefit. Without it, the system doesn't reliably affect behavior.
Transparency supports reciprocity. When my past actions in response to someone else's behavior are unclear, people who interact with me in the future will not be reliably affected by those actions.