Executive summary
If an agent finds water and then dies, can the others still reach it? The Valley turns that question into an experiment. A hundred language-model agents live and die in a shared physical world, and four such worlds are followed for thirty simulated days. The study tests whether knowledge, coordinated work and recovery after a death belong to the population, beyond any single agent. A system built to run for years will lose and replace parts along the way, and the Valley measures whether its capability survives that.
Findings
- Standard evaluations reset the world and stop at the score. An autonomous system acts on a world its own earlier choices have shaped, mistakes included, so its capability has to be measured over that history.
- Capability belongs to the system. The same model is more or less capable as its memory, communication and recovery change, and a population can finish work that no single member could complete in one sitting.
- Knowledge survives a death only through someone who learned it or a record that lasted. A discovery counts as shared when other agents can still act on it after its finder is gone.
- Only physical outcomes count. A confident report of finding water scores nothing until someone reaches the water.
For Durable Intelligence, the Valley takes the durability question from a single agent to a whole population.