Independent Measurement Under Policy Capture in Multi-Agent AI Systems: The Reserve-Measurement Hypothesis and a Testable Evaluation Framework

Status: Preprint / position-and-protocol paper.Empirical status: The proposed RMH experiment has not been conducted as part of this publication.Reuse: Further theoretical development, replication, falsification, experimental implementation, and derivative research are explicitly encouraged under the CC BY 4.0 licence.AI assistance disclosure: Microsoft Copilot was used for iterative drafting, synthesis, document production, and literature-assisted refinement. Mistral and SVEAGPT were used as adversarial reviewers during development. The author selected, evaluated, and adjudicated revisions. AI review is disclosed as provenance and is not treated as empirical evidence supporting RMH.