Relying on information from other people (social testimony) is essential for efficiently learning and reasoning about the world. However, determining who to trust is often challenging. In this paper, we argue that trust in social agents (i.e., those providing testimony) can be evaluated by assessing how optimally they have acquired their knowledge. Building on theories that describe knowledge acquisition as a stochastic search through a space of hypotheses, we present a framework which yields predictions about which agents will provide better testimony (because they are more likely to have uncovered higher-probability hypotheses) in different contexts. This approach allows us to jointly predict how the quality of testimony is affected by 1) features of the agents themselves, like their expertise; 2) consensus among multiple agents; and 3) features of the topic and hypothesis space, like its knowability. We present initial simulations demonstrating how even a basic implementation of our framework yields insight into which types of agents and topics are more likely to result in accurate testimony (and why). We conclude by discussing how this preliminary research might be extended to address more complicated social reasoning scenarios.