Active Metastability in Political Polarization: Out-of-Sample RCR Validation (1994–2024) and Forward Projection to 2028

Standard models of opinion dynamics — DeGroot consensus, bounded confidence, and the Voter model — predict convergence under conditions that should drive minority views toward assimilation. They fail to explain the empirical trajectory of US political polarization: a stable bimodal distribution, with the liberal minority (~28–35% of the electorate) persisting and growing over three decades while the moderate center has contracted from 43% to 33% of respondents. We apply the Recursion-Collapse-Recombination (RCR) framework to this persistence problem. RCR Model 2, initialized from the 1994 Pew Research near-unimodal baseline and run forward 30 steps under frozen canonical parameters, predicts the 2024 bimodal distribution with a raw bimodality index error of 0.117 — compared to 0.351–0.430 for the best social science baselines — a 3× reduction in out-of-sample prediction error. Global sensitivity analysis (Sobol indices, corrected estimator) identifies χ, the global collapse strength parameter, as the only driver with a statistically defensible first-order effect on Active Metastability variance (S₁ = 0.29, 95% CI [0.16, 0.45]). We interpret χ as institutional consensus pressure: the structural claim is that polarization deepens as consensus mechanisms weaken, not primarily as partisan identities intensify. RCR Model 5 (endogenous stochasticity) generates a forward projection from March 2026 through November 2028, predicting bimodality persistence above BI = 0.5 through both election cycles under the k=2 two-cluster geometry (95% CI lower bound 0.603 at Nov 2028; k=3 verification yields final BI = 0.377 with qualitative predictions preserved), volatility spikes of 2.5× baseline at each election event, and convergence to the 2024 empirical anchor by the 2028 presidential election. These predictions are falsifiable in real time as Pew and Gallup tracking data arrive.