Cosmological Boundaries Without a Global Initializer: Boundary Conditions, Measure, and Causal-Order Sum as One Declaration-Relative, Self-Uncertifiable History-Amplitude Package, Not an Externally Initialized Universe

This paper is the thirty-seventh in a sequence developing an interpretive framework that redescribes physical reality as a causally consistent history of information updates. It is the capstone panel (third of three) of the tenth grouping — the dynamical quantum-mechanics–gravity interface, or emergent-spacetime, trilogy (35 Emergent Spacetime → 36 Gravitational Dynamics → 37 Cosmological Boundaries) — and it closes that grouping, under one discipline it never violates: adoption and typing, not derivation or selection. When geometry is a quantum degree of freedom (Paper 32) reconstructed only from records (Paper 35) and its coupling to matter is a declaration-relative equation of state of records (Paper 36), one authority still stands at the interface: a Global Initializer that either sets the cosmological boundary condition from outside the history — the Past Hypothesis, the low-entropy boundary, "the wavefunction of the universe" — or presupposes that one uniquely correct cosmological measure exists. The thesis is that there is no Global Initializer relative to the framework's declared architecture. Wheeler–DeWitt's constraint, the Hartle–Hawking no-boundary and Vilenkin tunneling proposals, Page–Wootters relational time, and the Boltzmann-brain / cosmological-measure pathology are imported and typed, never re-derived, adjudicated, or solved. The framework types the whole cosmological object as one adopted, declaration-relative, self-uncertifiable history-amplitude specification package, bundling five modules the corpus already owns — the geometry sum (Paper 32), the causal-order sum (Paper 34), Paper 36's declared dynamical coupling, an adopted boundary condition (Paper 9, cosmologized), and a declared measure (Paper 11) — under the Wheeler–DeWitt constraint and relational-clock conditioning (Paper 8). The package is typed precisely: its amplitude component is physics, the measure is a specification that reads weights and self-locating credences off it, and the non-self-undermining condition is a meta-condition over the package. The foil is disjunctive, so P1 (No Global Initializer) retires each face separately: the boundary-setter face is displaced (the boundary enters only as a declared slot, and the constraint underdetermines it — Hartle–Hawking and Vilenkin both admissible, opposite-sign weights), the canonical-measure face is undercut (no single natural count; no self-certification), and the Branch-Decomposer sub-foil is declined (coherence is not identity; no canonical carving). R-C: cosmological boundary conditions are adopted, not derived (the Paper-9 lift; privileging one proposal as constraint-forced or self-certifying is the foil, but a principled, warrant-earning adoption is admissible). R-D: the cosmological measure is a typed specification, not a canonical count (the Paper-11 lift; a measure whose verdicts on invariant observables depend on the branch decomposition is inadmissible, while a declared non-observable decomposition is admissible-but-warrant-divergent). R-E: the package must pass the non-self-undermining filter — a cosmology whose measure withholds entitlement from its own supporting evidence is cognitively unstable, not false (the quantum-cosmology form of the Boltzmann-brain pathology; necessary but not sufficient, and high provenance prevalence does not suffice). R-A (flagship): on an explicit finite decoherent-history model, two specifications induce the same coarse-grained history-weight profile yet differ in warranted cosmological standing, because the branch/boundary decomposition is a declared, non-observable component — there is No Global Branch-Decomposer. R-F: no cosmology self-certifies. The Global Initializer is retired relative to the framework's declared architecture (the serious realist cousin — one true boundary, measure, or decomposition — is declined, not refuted); P1 is architecture-relative, not an ontological negati