We formulate and apply a machine- and domain-independent instability operator ∆Φ designed to detect transitions between three generic dynamical regimes: (i) a stable baseline regime (isostasis), (ii) an adaptive metastable regime (allostasis / instability), and (iii) a critical collapse regime. The operator is constructed as a weighted aggregation of deviations along three independent axes: structural (S), informational (I), and coherence-related (C), with α+β+γ=1. We first summarize an independent, fully reproducible EEG validation on public PhysioNet data (including CHB–MIT Scalp EEG), where elevated ∆Φ excursions precede seizure onset and yield strong predictive performance in a windowed pipeline. We then show that the same operator admits a natural instantiation in magnetically confined tokamak plasmas (JET / DIII-D / ASDEX Upgrade) using standard diagnostics, and we provide operational mapping tables and expected ∆Φ(t) families for L–H transitions, ELM-like bursts, and major disruptions. Crucially, the operator definition is kept fixed across domains and is not retrained or reparameterized, enabling direct falsification through archival tests. This positions ∆Φ as a universal instability indicator for open, non-equilibrium systems rather than a domain-specific classifier.