A classification of theoretical-term reference and its change across theory succession, not a verdict on which terms in fact refer. The corpus fixes what a record is of (Paper 13), how observers co-refer and translate (Paper 14), when structure is objective (Paper 15) or recoverable (Paper 19), what is true given reference (Paper 20), and how inquiry revises (Paper 22) — but nowhere types what a theoretical term refers to inside a history, or the referential relation between two theories across a succession T -> T'. While no such layer exists, a Global Semanticist — an external meaning-fixer who supplies reference from outside the history — works by default, and naturalized incommensurability has no honest home. This paper types that layer. Three type facts stand in print and bind the paper: "Typed is not adopted; classified is not answered. " / "A term-reference status is a type, not a world-side verdict. " / "A cross-theory reference relation is a relation, not a translation. " The construction is a term-reference grounding object U (t) = (Dᵤse, Xₜ, Piₜ, gammaₜ) — a theoretical term's calibrated-usage distribution, its declared world-target commitments, a provenance graph of introductions and re-fixings, and a term-individuation grain — with a normal form (TermRefSpec) and a barred-fields box. It is distinct from Paper 13's per-token lineage: Paper 13 is per-token, whereas U (t) aggregates across tokens, observers, and time and handles usage drift and re-fixing without an external averaging authority. On it, TermRefStatus in typed-refers, typed-reference-failure, typed-indeterminate, typed-partial — a status type, never a world-side verdict — is assigned by a decidable four-check typing test in which admission (declaration) is sharply separated from routing (emptiness), governed by two declared parameters, a dominance threshold thetadom in 1/2, 1) (default 1/2, strict majority) and a non-negligible-weight threshold thetaₙn in (0, 1/2; the routing is a total, disjoint, decidable algorithm over the finite weighted incompatibility graph. Reference change is typed by TheoryRefRel over T -> T' in co-refer, partial-co-reference, reference-shift, reference-loss/gain, talk-past, indeterminate — six cells on set-valued class memberships — with fixed evaluation inputs and a disjointness lemma (Lemma D) separating it from Paper 14's inter-observer translation — a relation, not a map. Schematic cases (a caloric-type, a phlogiston-type, a continuant-type term) are typed by displayed usage-profile, not by historical reputation. The ceiling is declared and patrolled: typing is not adopting; classifying is not answering. The substantive which-terms-in-fact-refer question is declined, not refuted; theory-whole referential content, truth, realism, revision dynamics, objectivity, recovery, and inter-observer translation all stay at their owners' seats; strict recovery of Paper 13's per-record semantics is exact (Lemma R-th; not up to paraphrase or intertranslation, which would be K0). The predicted landing is LayerStatus (theoretical-term reference) = scoped, with a null via K0 an honest branch and a Semanticist-dependence theorem (K3) declared and patrolled. Nothing is adopted; no truth, objectivity, or world-side reference is claimed. The paper engages the philosophy-of-reference literature (Putnam 1975; Kripke 1980; Field 1973; Kitcher 1978; Kuhn 1962; Feyerabend 1962; Worrall 1989; Laudan 1981) as display and boundary only, naturalizing the causal-historical chain as the provenance graph, typing Field's partial reference as the typed-partial cell, localizing Kuhn/Feyerabend incommensurability to the talk-past cell, bounding Worrall's structural realism as Paper 20's territory, and supplying the typed vocabulary Laudan's meta-induction presupposes without issuing any first-order verdict.
Tomoyuki Uchida (Wed,) studied this question.