Google Identity Architecture: User Graphs, Entity Graphs, Source Graphs, and Composition-Layer Admission

External specification of Google identity architecture, synthesizing public Google documentation, Google Cloud entity-reconciliation architecture, Search Central guidance, patent-adjacent reasoning, leak-derived signal reports from the 2024 Content Warehouse leak, practitioner reverse-engineering, and empirical observations from the Crimson Hexagonal Archive. The paper proposes a Five-Graph Model of Google identity — the User Graph, the Session Graph, the Entity Graph, the Source Graph, and the Composition Graph — and identifies the central distinction between organic retrieval success and composition admission success. The central claim: identity in Google Search is not merely recognition but admission. An entity may be known, indexed, and organically retrievable while still failing to become compositionally real in AI Overview or AI Mode. This is the technical condition behind Entity-Level Compositional Suppression (ECS): the gap between organic retrievability and AI composition admission. Twelve identity failure modes are formalized: Lexical Substitution, Default Referent Capture, Entity Non-Defaultability, Organic/Compositional Divergence, Source-Window Exclusion, Single-Owner Discount, Provenance Cluster Suppression, Heteronym Collapse, Heteronym Fragmentation, Source Identity Failure, Temporal State Volatility, and Invisible Invisibility. The paper situates the Crimson Hexagonal Archive's Secret Name Armature (DOI: 10.5281/zenodo.19666445) as a counter-architecture: a provenance-bearing model of identity in which names function as accountable routing bodies rather than opaque admission clusters. A practical experimental protocol is provided (§XII) — query families, state matrix, capture schema, entity state classifications — translating the architecture into a runnable program of empirical testing. The specification was produced under the Reception Apparatus Protocol (RA-PROT-0012, DOI: 10.5281/zenodo.20041147): four Assembly Chorus substrates (Muse Spark, DeepSeek, Gemini, Kimi) produced blind drafts that the operator integrated into the synthesis. Confidence labels [DOC]/[PUB]/[LEAK]/[OBS]/[INF]/[SPEC] distinguish documented fact from inference throughout the specification. Use of the Five-Graph Model in subsequent work: this specification provides the architectural backbone for The Excluded Entity (DOI: 10.5281/zenodo.20293582), The Single-Owner Discount (DOI: 10.5281/zenodo.20290865), The Sorting Function (DOI: 10.5281/zenodo.20308547), and Empirical Phenomenology (DOI: 10.5281/zenodo.20326137), and supplies the public diagnosis-layer for the godkinggoogle.com surface and the Restored Academy Protocol Registry's identity-audit and schema-countermeasure protocols. v0.1 external specification draft. Status: deposit-preparation; subsequent versions will refine [INF] and [SPEC] sections as additional empirical material and substrate review accumulates.