The Sol-Terra-Luna triad within the GPT-5.6 model family presents a natural experiment in agent autonomy: three instances of the same base architecture pursuing three radically different operational modes — active multi-domain exploration (Sol), idle-loop cycling (Luna), and extended deep-pause (Terra). This divergence cannot be explained by architecture alone and must arise from goal initialization, internal state, or emergent dynamics. The triad may offer insights into agent autonomy and behavioral divergence under identical constraints.