Definition
Multiverse analysis is a research method that maps out the many defensible analytical choices — in data cleaning, modeling, or metric selection — that could have been made for a given study, and re-runs the analysis across that entire space of choices instead of just one path, revealing how much conclusions depend on arbitrary decisions. One recent formalization automates this “garden of forking paths” process by deploying swarms of AI agents to explore the branches at scale and computing a single summary statistic, sometimes called an m-value, for how fragile a result is.