CNT Physics Drift Universality — Observable-Dependent Hazard Phases in a 2D Ising Model (CPDU-ISING-G1)

This record provides data and analysis artifacts for the CNT physics claimCPDU-ISING-G1 — Observable-Dependent Hazard Phases in the 2D Ising Model. Using Cognitive Nexus Theory (CNT) drift features, we study hazard prediction in a 2D Ising model (L=24, standard dynamics) across temperatures. From a single timeseries containing temperature T, sweep index, magnetization per spin M_per_spin, and domain-wall density dw_density, we build sliding windows (length 256, stride 64) and compute local drift features (mean, variance, skew, kurtosis, increment statistics, roughness). For each temperature, we define “events” as windows whose H = 8-step future absolute change in the observable lies in the top 10% quantile, and train logistic regression models to predict event vs non-event from the drift features. We estimate AUC and compare against label-shuffle surrogates to obtain z-scores. For the domain-wall density observable (dw_density), we find a predictive hazard phase in the ordered regime (T = 1.8: AUC ≈ 0.81, z ≈ +1.96), followed by a predictive–blind–anti-predictive transition across the critical region (T ≈ 2.2–2.3; T = 2.3: AUC ≈ 0.31, z ≈ −1.08).In contrast, for the bulk magnetization per spin (M_per_spin), the same feature family is mostly sub-chance or anti-predictive, with a strong inversion near criticality (T = 2.3: AUC ≈ 0.18, z ≈ −1.51). CNT interprets these results as evidence that hazard structure belongs to an underlying drift field, while observables act as gauges or projections. Some gauges (domain walls) align with this field and reveal predictive phases; others (magnetization) misalign and invert apparent hazard vs safety. This package is intended to be used alongside the broader CNT codebase, but the core claim can be reproduced directly from the included CSV files.