Global Term Key: Identity Primitives in Governed Knowledge Systems

Knowledge systems often use a surface word, embedding neighborhood, graph node, URI, or schema type as though it were already a stable semantic identity. These devices are valuable, but none alone decides whether two uses carry the same binding, whether one term names several bindings, or whether a binding may be used for a consequential claim. When similarity, co-occurrence, or shared spelling is silently promoted to identity, concept drift, hub gravity, semantic smearing, and polysemy-driven conflation follow. This paper develops the Global Term Key (GTK) as a governed lexical interface around a disambiguated semantic binding. Its original three-part insight remains: a GTK exposes (1) the active binding identity, (2) the aliases admitted for that binding, and (3) the competing bindings reachable from the same surface forms. Corpus integration sharpens the architecture. The coordinate that locates a binding belongs to the Semantic Coordinate Identity Layer at Representation-Pointer Zoom 0 (CSIL/RPZ0). GTK lives at RPZ1: it supplies the governed bound-term, named renderings, aliases, inflections, and competing-binding fan through which that coordinate becomes humanly and computationally navigable. A disambiguation context D selects among candidates without converting probability or convenience into semantic authority. The invariant floor, definition, and admissible variation associated with a binding are therefore linked specification at RPZ2 and beyond, not all compressed into the key itself. Likewise, adjacency is not universally established by removing one alleged invariant. It is a typed, lens-relative relation that must state which features, ontology, metric, registry, authority, and evidence make two bindings neighbors. Content addressing preserves exact record identity under a declared schema; it does not prove that the record is semantically correct. GTK thereby becomes a bridge among lexical resolution, semantic coordinates, the Governed Meaning Base (GMB), semantic hashing, governed context formation, and machine-operable semantic languages. The proposal is architectural rather than empirical. It specifies a resolution contract, failure semantics, realization funnel, conformance questions, and counterevidence program. It does not claim that all concepts possess crisp necessary-and-sufficient invariants, that automated disambiguation is infallible, or that this paper establishes novelty relative to current lexical and knowledge-representation research.