100 LLM Agents Ran a Town Economy for 26 Weeks and Repriced 0.3% of 3,981 Menu Items After a 12x Demand Shock

arXiv 2609.11108·medium signalXBlueskyLinkedInCopy link

arXiv 2609.11108 (submitted 10 Sep 2026) put 100 memory-equipped LLM agents in charge of a closed, money-conserving economy on real Pokhara Lakeside geography for up to 26 simulated weeks, well past the 1-2 weeks typical of agent-society studies, across 91 validated runs totaling 2.44M decisions and 21.5B tokens. A 12x tourist demand shock raises business revenue 4.62x, but wages move 1.03x (p=0.42) and only 0.3% of menu items are ever repriced (p=0.47); a randomized NPR 5,000 cash transfer to 20 agents is 96.7% still held 311 pulses later, with marginal propensity to consume at 3-4%. Wealth-distribution correlation is 0.964 at two weeks but falls to 0.832 at 12 and 0.752 at 26, a horizon dependence no short study can observe.