Islands of Stability
Six Falsifiable Propositions on Where Complex Systems Work, and Their First Tests in AI Agents and Across the Sciences
Most ways a complex system could be run do not work. The ones that do occupy small, bounded regions of the settings available, islands of stability, and staying inside one often takes continuous effort. This paper states that idea as six propositions precise enough to be shown wrong, each naming the observation that would kill it, and reports their first tests, each fixed and deposited with a DOI before it ran. In 18,720 runs of three local AI agents, the working region was a slope along the failure axis, not a band with an edge, and a push revealed what fault-free success hid: after an injected failure the agents gave up in 15 to 61 percent of runs. Tests in a lake model, the last ice age, the business cycles of 42 countries and measles in 138 countries returned survives, restricted and unresolved. No registered test has killed a proposition. The rules, data and code of every test are public at islandsofstability.com.
