Classical epidemic models assume the pathogen doesn't change. The virus that infects patient 1 is the virus that infects patient 10,000. This makes the math tractable: exponential growth with fixed parameters, R₀ as a constant, interventions as simple multipliers on transmission.
Allow the pathogen to evolve — let infectiousness change across generations — and the dynamics break. Growth becomes superexponential (arXiv:2603.18801). Not just faster-than-expected but qualitatively different: the growth rate itself accelerates because the pathogen is becoming more transmissible while it's spreading. Two clocks — the epidemic clock (how fast it spreads) and the evolutionary clock (how fast it adapts) — are racing each other.
The devastating consequence: lifting interventions too early produces worse outcomes than never intervening at all. Intervention simultaneously reduces cases AND slows evolution. Remove it, and the evolved, more infectious variant encounters a largely susceptible population. The intervention created the selection pressure; its removal releases the adapted pathogen. The paradox is real: partial action is worse than inaction when the pathogen can learn from the pressure.
The structural insight is about timescale coupling. When evolutionary and epidemic timescales are well-separated (evolution much slower than transmission), the classical model works — the pathogen is effectively static. When they overlap, the system enters a regime where analytical expressions for the critical mutation rate and intervention timing govern outcomes. The answer to “when should we intervene?” depends not on the current R₀ but on how fast R₀ can change.