Measuring AI R&D Automation Before Intelligence Explosion: A Governance-Grade Metric Framework for Loop Closure, Oversight Latency, Verification Independence, and Authority Distance

This paper develops a governance-grade measurement framework for AI research and development automation under conditions in which increasingly autonomous AI systems may participate in, lead, or close progressively larger portions of the AI R&D loop. Existing measurements of AI R&D automation provide an important starting point, but a percentage of work performed by AI does not by itself reveal whether an automated research process is becoming structurally harder to govern. Two systems may exhibit similar levels of task automation while differing substantially in whether AI systems can select objectives, modify research infrastructure, influence evaluators, persist improvements, or convert successful experiments into changes to protected capability or authority. The paper therefore treats AI R&D automation as a control topology rather than only as a labor-automation percentage. It introduces a governance-grade metric vector covering Automation Coverage, Loop-Closure Depth, Self-Modification Depth, Human Intervention Distance, Control-Latency Ratio, Verification Independence, Evaluator Entanglement, Authority Distance, Protected-State Transition Rate, Rollback Distance, and Revocation Persistence. The framework distinguishes task automation from recursive loop closure and measures how many independent control boundaries remain between an AI-generated improvement and a protected state transition. It also develops denominator discipline, measurement protocols, trigger semantics, adversarial validation methods, cross-organization comparability requirements, evidence-lineage requirements, red-team tests, and reporting structures for governance use. The central proposition is that the governance significance of AI R&D automation depends not only on how much work AI performs, but on how far an automated improvement can travel through proposal, execution, evaluation, selection, persistence, and re-entry before encountering an independent human or technical boundary. The framework is designed to support cross-laboratory comparison without requiring publication of model weights, proprietary research details, confidential infrastructure parameters, or other sensitive implementation information. This publication is intentionally limited to public research-level abstractions and does not disclose unpublished patent claim language, confidential claim charts, private source locators, provider-specific production parameters, or non-public implementation details. Structural Paper Series - Paper 02 Final Publication Edition v2.1 Research Program on Deterministic Infrastructure and Human-Centered AI Coordination Transition Intelligence Institute, Switzerland Institutional Establishment in Preparation Foundational Research Signature: The Second Waters