Recursive Transport Topology Detects Structural Regime Transitions Associated with Behavioral Failure in World Models

Read full text
Authors: Jeffery Scott Allbright
Cited by: 1
References: 2

Abstract

Abstract World models are typically evaluated through prediction accuracy or downstream task success, yet these measures do not directly test whether an internal representation has crossed into a structurally unstable operating regime. We evaluate Recursive Indexing (RIX), a fixed-reference analysis of latent transport topology, in a pretrained LeWM PushT world model exposed to controlled Gaussian observation stress. Progressive stress produced a sharp nonlinear reorganization of latent neighborhoods, fixed-reference basins, transition pathways, and pathway entropy while conventional pointwise latent displacement remained small. The dominant structural breakpoint localized to the 0.05–0.10 stress interval across all 320 analytical configurations. In an independent closed-loop calibration, task success declined from 80% at stress 0.00 to 20% at 0.05 and 0% at 0.10 and above. These results support a regime-level interpretation of world-model reliability: latent representations can remain pointwise close while their transport organization has already changed substantially, and the resulting structural transition can align with behavioral collapse.

Fields of Study & Subjects

Functional Brain Connectivity Studies Embodied and Extended Cognition Mental Health Research Topics

Publication details

Journal / source
Research Square
Publisher
Springer Science and Business Media LLC
Publication date
10 July 2026
Document type
Preprint