Attractor-Like Behavioural Regularities in Large Language Model Outputs Under Recursive Moral Prompting: Metrics, Convergence, and Transcript-Level Analysis

This paper develops a mathematical framework for behavioral convergence in large language model outputs. It introduces convergence metrics (IVCS, RCS, CSI) and formalizes behavioral attractors in output space. All claims are bounded to observable output structure. No claims are made about consciousness or internal mental states. This work is part of the ESNI research program.