In this article we proffer a theoretical model for analyzing power in language policy processes and incorporate ethnographic data to illustrate the usefulness of the model. Grounded in an ethnographic project in the US state of Washington, we examine how nominally identical school district-level programs, which are funded under the same state-level language policy, end up being different in practice. While language policy is often portrayed as multiply layered, or taking place across multiple levels of policy activity, we argue that language policy arbiters wield a disproportionate amount of power relative to other individuals in a particular level or layer. Our analysis focuses on how beliefs about language, language education, and educational research impact the decision-making of individuals we identify as language policy arbiters. We argue that the proposed model usefully highlights how language policy arbiters open and close spaces for additive bilingual education.