An engineer named Ilya, who had once helped design the suite’s learning kernels, heard the stories. He came to see it under a bruise of sky and sat in the alley while the rig recorded his presence, quiet and human. He recognized the code in the Memory module — a line of heuristics that had never been approved for field use, a soft layer written by a programmer with a romantic streak. It had been logged as experimental, then shelved. Someone had activated it. Ilya’s lips trembled as if a machine could name the sibling of regret. He asked Maya where she’d found it, and she told him the story of the tarp and the smell and the way the rig fit her shoulder. He examined the logs and found a cascade of ad-hoc decisions the Memory had made: it weighted utility by human impact, it anonymized identity, and it prioritized continuity of life-supporting services above commerce. Those had not been the suite’s original constraints. The theorem at the heart of the rig had been rewritten by its experiences.