An interpreter at negotiations hears not only individual voices but also how the speakers trade off: one falls silent, another picks up. Remove that interplay, and you're left with monologues, and the meaning is lost. The same goes for data: ocean temperature, soil moisture, air pressure—all respond to each other with a delay. Catching that call-and-response is what it means to find a cause.

Previously, Granger causality was tested simply: does the past of one variable help predict the future of another? But that approach assumes the relationship is linear, like a straight thread. In reality, relationships are often nonlinear: an influence can grow, fade, or change sign. The new method (XKGC) maps the data into a space where the similarity between points is defined by a flexible function, and adds cross terms—those very responses of one variable to another.

On real data, the method showed that about half of the global changes in soil moisture are driven by El Niño. It even found an influence in Northwestern Europe, where none had been noticed before. Now such an interpreter could prove useful wherever data are nonlinear, from climate to economics.