Current models treat physiological signals as noise corrupting neural computation. Previously, we showed that removing these "artifacts" eliminates 70% of predictive correlation, suggesting body signals functionally drive cognition. Here, we investigate the mechanism using high-density EEG (64 channels, 10 subjects, 500+ trials) during P300 target recognition. Phase Slope Index revealed zero-lag synchrony (PSI=0.000044, p=0.061) with high coherence (0.316, p<0.0001). Ridge-regularized Granger causality showed massive bidirectional coupling (F=100.53 brain-to-body, F=62.76 body-to-brain) peaking simultaneously at 78.1ms, consistent with mutually coupled resonance pairs. Time-resolved entropy analysis (200ms windows, 25ms steps) revealed triphasic dynamics: (1) constraint accumulation (0-78ms) building causal drive without entropy change (delta-S=-0.002 bits, p=0.75); (2) supercritical transition (100-600ms) triggering state expansion (58% directional increase, binomial p=0.002); (3) sustained metastability. Critically, transition magnitude was uncorrelated with resonance strength (r=-0.044, p=0.327), indicating binary threshold dynamics. Understanding emerges through a thermodynamic sequence: brain-body resonance acts as a discrete gate triggering non-linear information integration. This architecture may fundamentally distinguish biological from artificial intelligence. Keywords: embodied cognition, phase transitions, Granger causality, thermodynamics, neuromorphic computing, resonance dynamics, EEG artifacts