The arXiv preprint (submitted April 2026) extends the deterministic Bertotti-Modanese 2018 framework by embedding two PDMP mechanisms: random audit jumps that push agents toward compliance and imitation-driven jumps that pull them toward evasion. Unlike the original static treatment of behavior, these processes conserve population and income totals while generating opposing attractors—full compliance under audits alone, full evasion under imitation alone. Their interaction produces bounded oscillations around the deterministic mean, hinting at convergence to a non-degenerate stationary distribution. This preprint remains theoretical; no empirical calibration or real taxpayer panel is used, only numerical simulations whose parameter ranges are not validated against observed audit rates or network data. The work misses policy translation: how audit frequency should scale with network density to dampen cycles, and whether the implied stationary distribution matches longitudinal tax-gap estimates from IRS or HMRC panels. Related literature includes Allingham-Sandmo (1972) expected-utility baseline, which treats evasion as a one-shot gamble without social feedback, and more recent agent-based studies (e.g., Hokamp & Pickhardt 2010) that already document cyclical compliance but lack the rigorous PDMP convergence proofs offered here. The advance lies in mathematically guaranteeing conservation laws while allowing endogenous volatility—yet the absence of empirical anchoring limits immediate applicability to enforcement design.