When Can Brain Connectivity Track the Working Mind? A Large-Scale Benchmark of Dynamic Functional Connectivity Across...

Dynamic functional connectivity (dFC) - the time-varying reconfiguration of brain network interactions - has become a widely adopted method for studying how neural dynamics reflect ongoing cognition. Yet a fundamental question remains unresolved: can dFC reliably track when a person is cognitively engaged, and if not, why does it fail? Here, we address this question through a large-scale benchmark of seven widely used dFC methods, evaluating how well each predicts task presence across 16 fMRI datasets encompassing over 1,500 participants and 28 distinct experimental settings, complemented by realistic simulated data. Across experimental data, dFC-based tracking of cognitive engagement was unreliable in many cases: most method-experiment combinations performed near chance, and no single method succeeded across all contexts. This failure, however, was not uniform. Both experimental and simulated data showed that decoding performance varied systematically with three interacting factors-experimental design, data quality, and the choice of dFC method - rather than depending on dFC features alone. Critically, we identify specific experimental design conditions associated with more reliable tracking: paradigms with longer, more regular task blocks and fewer task-rest transitions were substantially more decodable, while data quality independently influenced performance across methods. These findings offer actionable principles for when dFC can - and cannot - be expected to serve as a reliable marker of underlying cognitive states.