Applicability-First Evaluation Module 02: Memory, Context, Delegation, and Manual Boundary Integrity in AI Agents: Persistence Admissibility, Tool-Mediated Context Transfer, Manual Human Grants, and Local/Cloud Data Sovereignty in High-Consequence and Decision-Bearing Systems

This release is a separate doctrinal evaluation module related to the canonical baseline publication of the Applicability Boundary Doctrine, the earlier doctrinal extension on observation validity, interpretation cutoff, and bounded reality verification, the separate doctrinal specifications on epistemic applicability, the separate doctrinal specification on reality verification, the separate doctrinal specification on approval-execution separation, the separate doctrinal specification on substrate integrity, the separate doctrinal specification on human-institutional responsibility, the separate doctrinal specification on decorative governance, and the separate doctrinal evaluation module on vendor claim admissibility. This record should be read alongside the canonical Applicability Boundary Doctrine, DOI: 10.5281/zenodo.19425317, the earlier related doctrinal extension, DOI: 10.5281/zenodo.19443895, the separate doctrinal specification on epistemic applicability, DOI: 10.5281/zenodo.19447536, the separate doctrinal specification on reality verification, DOI: 10.5281/zenodo.19457414, the separate doctrinal specification on approval-execution separation, DOI: 10.5281/zenodo.19462291, the separate doctrinal specification on substrate integrity, DOI: 10.5281/zenodo.19463230, the separate doctrinal specification on human-institutional responsibility, DOI: 10.5281/zenodo.19476910, the separate doctrinal specification on decorative governance, DOI: 10.5281/zenodo.19478004, and the separate doctrinal evaluation module on vendor claim admissibility, DOI: 10.5281/zenodo.19487979.It is a separate record and not a version of any prior publication. It formalizes a distinct doctrinal question within the same authored framework: when do persistent memory, assembled context, delegated chains, computer-use reach, browser reach, manual grants, local execution, external transfer, bounded cloud invocation, and connected tools acquire defensible standing for deployment-grade, assurance-grade, or consequence-bearing use? This release should be understood as a specialized evaluation module within the already-published consequence chain and across the orthogonal human-institutional responsibility and decorative governance contours. In doctrinal order, substrate integrity, observation validity, epistemic applicability, reality verification, approval-execution separation, the applicability boundary, and continuation legitimacy remain sequential machine-side and consequence-side conditions; human-institutional responsibility and decorative governance remain orthogonal across them; this module specializes whether memory, context, delegation, manual boundary control, and user-data handling may legitimately acquire standing under that existing chain. The release focuses on high-consequence and decision-bearing settings in which AI agents may remain technically functional while persistent context, task continuity, tool metadata, local execution, remembered approvals, browser state, delegated handoffs, or cloud convenience are treated as if they conferred authority, inherited applicability, or consequence legitimacy. Its purpose is doctrinal and analytical. It is intended as a citable public reference for continued work on:- memory admissibility- persistent context and cross-session standing- context assembly and tool-mediated context transfer- delegation re-establishment- delegated applicability propagation risk- manual-origin distinguishability- perimeter-scoped access- user data sovereignty- local execution non-egress- point cloud invocation- output reconstruction risk- adversarial memory/context surfaces- test–attack distinction- model-initiated escalation- encryption without authority transfer This release deepens the authored doctrinal framework. It does not replace the canonical baseline publication, the earlier doctrinal extension, or the separate doctrinal specifications on epistemic applicability, reality verification, approval-execution separation, substrate integrity, human-institutional responsibility, decorative governance, or vendor claim admissibility. It has independent citation value as a separate doctrinal evaluation module within the same authored framework. No implementation guidance is provided in this release. No memory architecture, browser-control stack, runtime enforcement design, control map, scoring protocol, commercial evaluation system, operational method, deployment pattern, or integration recipe is disclosed. The publication is limited to doctrinal articulation, evidentiary framing, legitimacy analysis, boundary clarification, and a staged-adoption companion note that does not convert partial implementation into full admissibility. Derivative interpretation, structurally similar reframing, internal reconstruction, selective reuse, or downstream implementation based on this release does not inherit authorship, originality, canonical standing, or independent evidentiary standing. No patent license or implied commercial implementation right is granted by this release. Commercial implementation, operational integration, productization, assurance packaging, procurement use, certification use, or other consequence-bearing application of the framework requires separate written permission or license from the author unless independently established without reliance on this authored framework. Earlier public materials, the canonical baseline publication, the earlier doctrinal extension, the separate doctrinal specifications on epistemic applicability, reality verification, approval-execution separation, substrate integrity, human-institutional responsibility, decorative governance, vendor claim admissibility, and independently timestamped records support continuity of authorship and development history. Canonical doctrine site: https://gotocalmaidp.github.io/doctrine-site-box/