LSFM Analysis of Regime Transitions in HL-3 Tokamak

 Recent disruption prediction on the HL-3 tokamak (Yang et al., Nuclear Fusion 2025) revealed systematic drift across five operational regimes, necessitating Predict-First Neural Network (PFNN) architectures for reliable performance. We provide the first theoretical interpretation using the Log-Scaling Flow Metric (LSFM), χ(λ) = [2 ln(λ)] −1 , which quantifies deviation from optimal recursive scaling at λ * = e ≈ 2.718. Analysis of HL-3 parameter evolution suggests Regime III (lowest disruption ratio 15%) corresponds to χ ≈ 0.5, while other regimes show χ drift correlating with instability. We demonstrate PFNN's 9.6% AUC improvement emerges from implicit χ-tracking, explaining prediction of novel disruption types without training examples. Six falsifiable predictions with validation protocols are proposed, including real-time χ monitoring for disruption avoidance. This framework provides physics-based guidance for ITER commissioning under non-stationary conditions.