In federated distributed systems, such as the Internet and the public cloud, the constituent systems can differ in their configuration and provisioning, resulting in significant impacts on the performance, robustness, and security of applications. Yet these systems lack support for distinguishing such characteristics, resulting in uninformed service selection and poor inter-operator coordination. This thesis presents the design and implementation of a trustworthy knowledge plane that can determine such characteristics about autonomous networks on the Internet. A knowledge plane collects the state of network devices and participants. Using this state, applications infer whether a network possesses some characteristic of interest. The knowledge plane uses attestation to attribute state descriptions to the principals that generated them, thereby making the results of inference more trustworthy. Trustworthy knowledge planes enable applications to establish stronger assumptions about their network operating environment, resulting in improved robustness and reduced deployment barriers. We have prototyped the knowledge plane and associated devices. Experience with deploying analyses over production networks demonstrate that knowledge planes impose low cost and can scale to support Internet-scale networks.