Link-Native Distributed Computing: Computation as Constrained Traversal of a Hash-Addressed Graph

We present a formal model of distributed computation in which the sole primitive is the directed, hash-addressed link. In this model, computation is not performed by invoking functions on data, but by constraining and traversing a dynamic graph whose nodes carry activation vectors and whose edges carry semantic weights. State never exists globally; it is constituted locally through worker operators that observe graph neighborhoods, emit signals, and rewrite edges. We formalize the model, describe a five-layer worker taxonomy, define a composite semantic distance metric, and demonstrate that coordination, identity, computation, and visual projection collapse into a single underlying structure. We argue that this constitutes a novel computational paradigm — link-native computing — that is related to but distinct from the actor model, graph rewriting systems, content-addressable networks, and hypermedia-driven APIs.