MiroFish provides a workflow for graph-native predictive simulation. Users upload PDF, Markdown, or plain-text source material, then frame a question about how the scenario may evolve. The system extracts entities, motives, and factual anchors before building a relationship graph. Platform-native agents interact over multiple rounds across Twitter- and Reddit-style channels, modeling narrative spread, stakeholder reactions, and changing pressures. Generated reports summarize the trajectory, predicted developments, major risks, key actors, evidence lines, confidence signals, and a longer explanation of the outcome. Users can inspect the graph and simulation, interrogate the resulting world through ReportAgent, or interview individual characters directly. Scenario templates cover public opinion, product launches, policy reactions, finance, brand crises, literary continuation, and general forecasting.