Multi-agent systems driven by a single objective frequently exhibit mode collapse, where exploration diversity vanishes and the system converges to a single behavioral mode. We propose WillCore v2.0, a self-regulating framework that combines role-based scoring (STABLE / EXPLORE / GROW), variance-driven dynamic pressure, and closed-loop adaptive bias feedback to maintain multi-mode coexistence without manual tuning. The framework originates from D.S. Theory's Convergence Density Transition (CDT) model, where the hidden-to-visible ratio C = H/(V+H) governs phase transitions at the critical threshold C = 2/3. We map STABLE to low-tier consolidation (V), EXPLORE to cross-tier collision, and GROW to high-tier emergence (H). Empirical results across 10 independent runs show convergence to STABLE 19.6%, EXPLORE 19.3%, GROW 61.1% with variance under 1.2%. Linearized stability analysis yields spectral radius λ = 0.986, confirming the adaptive bias fixed point lies well within the basin of attraction (f'(b*) = 0.28, less than 0.7% of the instability threshold).
(Demian) et al. (Wed,) studied this question.