When Should a World Model Switch? Surprise-Gated State-Space Models for Hierarchical Events

A predictive world model faces a recurrent dilemma: preserve context while an event is stable, yet rapidly reorganize state when the current dynamics no longer apply. Most video models leave this decision implicit, and most event-boundary detectors mark changes without letting those changes alter the model's internal dynamics. We introduce $\text{HEG-VMamba}$, a multimodal latent world model that operationalizes Event Segmentation Theory as explicit predictive computation: $\text{calibrated predictive innovation}$ becomes an endogenous control signal, and a learned hierarchical $\text{event hazard}$ jointly (i) expands a selective state-space model's discretization step, attenuating stale history, and (ii) retrieves an input-conditioned $\text{schema}$ that initializes the next state. Counterfactual utility regularization teaches the gate to switch only when reorganization improves future prediction, and a learned prior supplies event hazards during open-loop rollout; we prove a monotone bound on the direct contribution of pre-boundary state that a fully open schema-mixture removes exactly. $\text{HEG-VMamba}$ improves controlled open-loop prediction by $-11.5\%$ ($5/5$ seeds, $p=0.007$) (OOD $+6.7$–$12.6\%$ in $5/5$ seeds across all four OOD families); on Kinetics-GEBD its internal gate reaches Rel. Dis. mean-F1 $0.760$ (zero-shot 5-seed ensemble on $17,911$ val clips with extracted features; per-seed $0.754\pm 0.032$) without boundary-supervised representation learning, and the same gate reduces open-loop prediction error on KTH Actions at recurring action-regime starts while remaining a detector at exogenous cuts. In naturalistic fMRI, endogenous innovation adds unique cross-validated predictivity of $\Delta r_{\mathrm{CV}}=+0.013$ ($p=0.005$; $15/16$ subjects) in a pre-specified event network beyond audiovisual features and raw innovation, replicating on an independent film ($\Delta r=+0.005$, $p=0.028$), while the gate's discrete boundaries dissociably track sensory-change cortex ($16/16$ subjects). Together, these results establish $\text{event-structured switching}$ as a principled mechanism for continuous world modeling, and show that its functional decomposition reflects the organization of the human brain.