This paper identifies the boundary gap — the structural condition where no entity governs and no entity possesses the information needed to verify codebook alignment between systems — and proves via the closure property that no computation within a fixed state space resolves it. We introduce null uncertainty as a third epistemic category beyond Knight's taxonomy, where the system structurally cannot perceive the absence of a variable, and prove a computable lower bound on the number of misaligned configurations guaranteed indistinguishable from alignment by any within-system observable (Proposition 3). The bound is a pigeonhole argument over Kowalski's forced quantization theorem: for realistic organizations it exceeds 2^800. We formalize the liquid hypergraph — a data structure whose edges exist only because measurements were made — and provide a protocol (four-facet gauge, five-column canonical claims, TCP-style handshake) that implements the measurement layer. Two synthetic demonstrations show the mechanism. Code, datasets, and protocol specification: https://w3c-context-graph-community-group.github.io/protocol/ Note: still a working paper draft.