Theories allow us to use accessible data to answer questions about other domains. In the initial stages of inquiry, a theory is often accepted based on its promise for extending our epistemic reach in this sense. Using theory to gain access to unobserverable phenomena poses an obvious risk of circularity: the theory specifies dependencies that hold between data and the target phenomena, and the data provide evidence when interpreted in light of the theory. How does the successful use of the theory to gain access support the theory itself? Demanding strong evidence at the outset, to even accept a theory as a starting point for inquiry, would be counter-productive. Detailed evidence for the theory can best be obtained by exploiting the theory in ongoing research. I will argue that physicists often accept a theory as the basis for research in this sense, and consider how they respond to the risk of circularity. The crucial question regards whether the fundamental assumptions of the theory can be subjected to further tests. These tests are needed to justify taking a theory to capture the fundamental quantities and physical laws, rather than being merely compatible with a given body of data. I will consider how these general questions are reflected in early universe cosmology. I will argue that there are two distinctive obstacles to testing inflationary cosmology: (1) lack of sufficient theoretical constraints ("anything goes"); (2) lack of independent observational and experimental access. The second point reflects our practical limitations. I take the first point, however, as grounds for drawing a distinction between inflation and earlier physical theories: eternal inflation, in particular, undercuts the possibility of subjecting the fundamental assumptions of the theory to further tests.
