SNT-LIFE v2: Coevolutionary Digital Organisms Under Operator-Constrained Survival Dynamics

We introduce SNT-LIFE v2 as a minimal digital organism framework in which adaptation arises not from reward maximization but from survival as a constraint satisfaction problem. Organisms operate under a Survival Viability Condition (SVC) —a formal requirement that metabolic energy never falls below a lethal threshold—eliminating the reward function entirely and placing the system within viability theory rather than reinforcement learning. The framework employs a fixed, empirically grounded seven-operator algebra derived from the C. elegans connectome, which forms a closed Lie basis (Frobenius residual 4. 82 10^-15). Adaptation is confined to distributional shifts over this fixed operator set, making all behavioral change interpretable as reweighting of existing primitives rather than structural novelty. We demonstrate that coevolutionary pressure is both necessary and sufficient to sustain adaptive dynamics. In a predator-prey coevolution setting (N = 100 prey, 5 predators), populations exhibit strong arms-race coupling (r = -0. 983, p < 10^-21) across five independent seeds. By contrast, prey-only evolution under identical conditions produces near-zero gain (3\% over 50 generations). This Sim1/Sim2 contrast isolates coevolution as the causal driver. Large effect sizes are observed: SCC organisms outperform random baselines by d = 4. 06 and simple rule-based agents by d = 1. 42 (both p < 10^-19). Operator importance is interaction-dependent: single-operator ablation shows small effects (0-2\%), while coevolution-level ablation reveals critical roles for evasion and gating operators, with removal reducing arms-race coupling by r 0. 05-0. 06. The system exhibits Red Queen dynamics—phase-shifted oscillations with no monotonic improvement—driven by a closed ecological feedback loop: predator speed increase prey shift to anticipatory evasion predator overshoot predator speed decrease prey recovery. This frequency-dependent selection mechanism is identifiable at the operator level and emerges without any reward function.