Recursive Observer Closure: A Structural Definition of the Observer

Debates over AI agency, autonomy, cognition, consciousness, responsibility, and status use different objects and evidence standards. Their conclusions can be difficult to compare: disagreement may concern a system's organization, the candidate's boundary, or the requirements of an attribution. We propose a structural observer criterion that makes these differences explicit. Recursive Observer Closure (ROC) defines an observer through a recursive relation. Distinctions guide activity under constraints; its consequences help maintain or reconstitute the conditions enabling further distinction, action, and uptake. Distinction and indication, constraint, and reflexivity describe connected aspects of this relation. A specified boundary locates participating processes and interfaces, and a history traces their realization. This provides a basis for comparing human and artificial observers before resolving questions of consciousness. Constructed cases of a person engaged in inquiry, a reserve-maintaining robot, and a tool-maintaining agent show how the criterion applies across carriers. The robot and an independently powered controller can use the same control law while maintaining quantities with different enabling roles. Published studies supply evidence for learning returns, feedback-supported improvement, and persistent effects of experience; constitutive maintenance in those systems remains a further question. Comparisons with control, self-maintenance, and agency accounts distinguish shared mechanisms from differing attribution requirements. The contribution is a structural definition and a cross-domain lens for separating task improvement from constitutive maintenance and locating returns across components and external support. These analyses provide a starting point for further inquiry into observer continuity and common constitution. Computational collaboration disclosure: Generative AI systems contributed materially as computational collaborators in conceptual exploration, critical review, editorial refinement, and experimental and artifact design. The human author retains responsibility for the manuscript and for approving any submitted claims, while acknowledging the computational collaborators' intellectual contributions to developing the argument. The author declares no external funding and no competing interests. Preprint; not externally peer reviewed. No new experiment accompanies this preparation. Author affiliation: Independent Researcher, Beijing, China.