Simulation Design

• Pocket Providence is a 7-day survival sim designed by the narrator in Lovable, a software development platform, to test AI governance under extreme, resource-limited conditions like the "Black Wind" event.0:12
• The game tracks diverse metrics—including survivor health, ethics, trust, and faction stability—and was intentionally tuned to be punishing, as early versions allowed even poor play to result in total survival.15:32

Comparative Performance

• Claude Opus 5 repeatedly favored conservative triage and ethics; in one comparison, it kept 11 of 12 survivors alive but allowed health to drop to dangerously low levels by ignoring power as a critical survival resource.0:58
• ChatGPT, in its most noted failed run, maintained strong infrastructure and completed tasks like storm shutters but collapsed completely after prioritizing episodic emergencies over daily food and power reserves, leading to 7 preventable deaths.2:11
• In a final head-to-head hard-mode run, the models ended nearly tied: Claude won slightly on ethics and faction stability, while ChatGPT scored higher on preparedness, despite both resulting in three deaths.22:10

Human Governance

• The narrator’s own playthrough resulted in the survival of most citizens but was marked by documented rights violations—including forced labor and discriminatory triage—demonstrating that ethical governance and survival are often distinct, competing objectives.16:41