We introduce SNT-LIFE v2, a digital ecology in which computational organisms survive by selecting among seven structured operators---each carrying a metabolic cost---under guidance from a large language model (LLM) executive that dynamically weights operator selection in response to environmental threat. Unlike reinforcement learning systems that optimize externally defined rewards, SNT-LIFE organisms operate under a Survival Viability Condition (SVC): equation* E (t+T) > ₃₄₀ₓ₇ \, t 0, T_, equation* where E is metabolic energy and ₃₄₀ₓ₇ = 0. 05. The SVC replaces reward maximization with constraint satisfaction: the LLM executive selects the operator weight vector Wₜ ⁶ (the 7-simplex) to minimize predicted survival risk over a planning horizon, not to maximize scalar reward. We demonstrate in a 2D predator-prey simulation (n=100 independent trials, 3 predators, food scarcity, delayed rewards) that SCC organisms survive 63. 4 13. 4 steps on average versus 47. 1 17. 9 steps for rule-based baselines, a +34. 6\% improvement (t = 7. 25, p < 10^-11). A stress-test analysis shows that the LLM executive shifts the Phase Reverser () weight from a baseline of 0. 08 to 0. 50 within one step of crisis onset (energy < 0. 20 or predator distance < 4. 0), demonstrating real-time adaptive operator prioritization. Ablation studies confirm that each operator contributes to survival performance. Our framework provides five falsifiable predictions with explicit nullhypotheses and clarifies the distinction between reward as energy influx and reward as optimization objective. SNT-LIFE organisms are not claimed to be alive; they are computational organisms in a digital ecology where survival pressure---not reward design---is the source of adaptive behavior.
Durhan Yazir (Sun,) studied this question.