Hidden Pseudoreplication in Disorder-Averaged Transport Simulations on Parcellated Connectomes

Standard brain parcellation atlases assign identical region labels to all subjects, creating a hidden statistical dependency when diagonal site disorder is generated from node identity. We demonstrate that, in GKSL/Lindblad transport simulations on Human Connectome Project connectomes (5 subjects, Desikan-Killiany atlas), shared-disorder seeding produces intra-class correlations up to ICC = 0.74 on a 2-target basal ganglia pathway, reducing the effective sample size from 60 to approximately 15. A one-line fix—hashing the base seed with subject identity—eliminates the correlation. We propose a hybrid experimental design and recommend reporting ICC for all disorder-averaged connectome studies. Companion paper: "Noise-Assisted Transport Windows in Human Connectome Subgraphs" (doi:10.5281/zenodo.18519173).