inZOR-ND: Discovery of a Machine-Agnostic Empirical Law Candidate for Disruption Proximity from Plasma Current Dynamics

This work investigates whether a simple machine-agnostic indicator of disruption proximity can be derived from plasma current dynamics. Using data from multiple tokamaks (MAST, C-Mod and HL-2A) together with evolutionary exploration using the inZOR-ND framework, we identify a compact empirical relation linking disruption proximity to statistical properties of the plasma current signal. The resulting candidate relation is: η ≈ −kurtosis + 0.80·rate_ratio + 0.65·cv_ratio The formula combines a structural descriptor of the plasma current signal (kurtosis) with two complementary instability channels derived from fluctuation dynamics. Extensive tests were performed to evaluate the stability and generalization of the relation, including: cross-machine validation across multiple tokamaks leave-one-machine-out (LOMO) evaluation bootstrap robustness analysis coefficient sensitivity tests baseline ablation analysis temporal evolution analysis before disruption cross-machine collapse test of normalized trajectories The results show that the candidate relation: consistently separates disruptive from non-disruptive plasma states remains stable under resampling and coefficient perturbations generalizes across machines without retraining exhibits a consistent temporal evolution approaching disruption After per-machine normalization, the temporal trajectories collapse onto a common curve across tokamaks, suggesting a shared statistical structure of disruption proximity in plasma current dynamics. Rather than proposing a predictive black-box model, this study focuses on identifying simple structural relations that emerge consistently across machines. The results suggest that disruption proximity may be associated with a statistical transition in the plasma current signal characterized by loss of structural sharpness (kurtosis decrease) combined with amplification of instability dynamics. The discovered relation is therefore proposed as a machine-agnostic empirical law candidate for disruption proximity derived solely from plasma current dynamics. https://dumitrunovic-svg.github.io/inZOR-ND/index.html