Source code for real-time control systems often intertwines several concerns such as functionality, data flow, control flow, synchronization, timing, and architectural style. This combination of concerns makes software harder to write correctly, harder to verify, and harder to reuse. This paper proposes an execution architecture that makes such systems more analyzable, verifiable, and reusable by separating "essential code" (software specific to the physical platform, the physical environment, and mission goals) from "incidental code" (all other software, particularly architectural support software for combining together essential components). This architecture elevates two forms of processing as firstclass items: individual transformations of global state, as defined in pure functions, and rules of interaction of transformations, as managed by an engine that maintains certain invariants. Importantly, the explicit specification of these two forms of processing by systems engineers reduces sources of ambiguity in requirements.