This paper extends the Latent Risk Geometry Classification (LRGC) framework by introducing a quantitative causal tracing layer for measuring non-interchangeability between latent-risk geometries. Using frozen metrics and identical computational substrates, the study applies neural causal tracing to four previously defined geometries: Corridor Fragility (T-dominant), Hidden Scarring (M-dominant), Symmetry-Protected Binary Failure (S-dominant), and Visible Lock-In (R-dominant).Across independently trained Graph Convolutional Networks, the framework measures Interaction Ratio (IR), Causal Field Strength (FS), Field Coherence, and Redundancy Threshold under controlled do-interventions. The results demonstrate that the four geometries produce opposite causal interaction structures: synchronization-driven systems are super-additive and cascade-amplifying, while topology-, memory-, and reinforcement-dominant systems remain sub-additive with distinct spatial signatures.The study further connects these computational signatures to real-world infrastructure evidence, including ERCOT WESTEX transmission corridors and Con Edison utility-scale outage data. The work transforms LRGC from a qualitative taxonomy into a measurable causal diagnostic framework for infrastructure resilience and latent-risk assessment.
Thomas S. Mitchell (Sun,) studied this question.