To address the need for social experiments, such as public opinion interventions, that are difficult to conduct in reality, existing large-scale social simulation platforms face challenges like loss of focus in long-term deduction and difficulty for researchers to make mid-course adjustments. In response, teams from multiple universities have launched SocioVerse2, expanding simulations from cross-sectional surveys to intervenable longitudinal processes. The platform constructs a tree-like research framework consisting of a longitudinal simulation loop, a controllable research loop, and a social science agent infrastructure. It generates simulation branches with shared historical precursors through a replay mechanism to accurately attribute intervention effects. Meanwhile, it grants researchers full-process editing rights and provides standardized real-world population and environmental data services. Verified through multiple case studies, the platform effectively supports the reproduction of classical behavioral experiments, policy scenario simulations, and prediction of unknown issues, offering researchers an open-source infrastructure and online workbench for collaborative social experiments between humans and agents.