CRN Stage-II Dataset: Transport-derived decision bias on 1064 HCP connectomes

Reproducible dataset and analysis pipeline for the closed-loop architecture linking GKSL open-system transport on structural connectomes to neural decision dynamics. The bundle contains subject-level results for 1064 Human Connectome Project (HCP) participants (Lausanne scale-125, 234 regions). Stage-I computes a unified scalar interface A = ln(R₀*/R₀_cls) for each connectome, where R₀* is the peak selectivity ratio at optimal dephasing and R₀_cls is the classical (high-dephasing) baseline. Stage-II tests whether A biases decision outcomes in two independent commitment models: a zero-drift diffusion model (DDM, v = 0, pure prior) and a biophysical Wong–Wang attractor network (c' = 0, no sensory coherence). Key results: (i) Group-level paired advantage in both models (DDM: Wilcoxon z = −8.79 for RT, z = 27.80 for accuracy; WW: z = −28.07 for both; all p < 10⁻¹⁷). (ii) Dose-response: inter-subject variability in A predicts decision outcomes (WW: Pearson r(A, ΔRT) = −0.645, Spearman ρ = −0.610; permutation p ≤ 10⁻⁴). (iii) Topology disruption (Maslov–Sneppen rewiring) attenuates or eliminates the effect. (iv) Spatial density routing is falsified as a mechanism; the operative interface is flux-based. The dataset includes: screening input (fullgraph_screening_1064.csv), unified interface values, subject-level DDM and WW outputs, sensitivity analyses (β, drift, α, I₀, c'), alternative mapping controls (start-point vs drift vs boundary), RT quantile profiles, correlation permutation tests (H1 paired + H2 dose-response), topology disruption controls, and mechanistic diagnostics (spatial routing falsification, flux analysis). Two self-contained Python scripts reproduce all derived outputs from the screening CSV using only NumPy. All transport computations use the canonical CRN architecture: H = −γ·L_sym (symmetrised graph Laplacian), explicit sink states, accumulated probability via matrix exponential, applied to raw graphml connectomes with log1p normalisation. GKSL/Lindblad dynamics serve as a functional proxy for wave-like exploration; no claim of microscopic quantum coherence in brain tissue is made. Related deposits: HCP reproducibility bundle (doi:10.5281/zenodo.18519173), Drosophila mushroom body dataset (doi:10.5281/zenodo.18697116), Mouse cortex applicability boundary (doi:10.5281/zenodo.18968563).