Runtime Characterization of Prepared Multi-State Photogrammetric Assets in Unreal Engine 5

Interactive use of prepared multi-state photogrammetric assets raises separate questions about local state propagation, rendered response, runtime cost and restoration of saved scene state. We characterize the retained ODD INN implementation through native graph inspection and instrumented Unreal Engine 5.8.3 execution on macOS/Metal. Four trajectories using two assets yielded 36 dependent observations with exact selected-component scalar propagation and unchanged unselected component payloads. Recorded BaseColor and SceneDepth buffers repeated at revisited inputs and between corresponding processes. Inputs 0.75 and 1.0 produced identical recorded buffers; small repeatable color differences also occurred in the unselected screen region. A 75-process timing experiment compared five recipient counts and three workloads. Across 1–40 recipients, the median of five process medians of central processing unit (CPU) time within the repeated-input invocation scope increased from 0.0545 to 0.4376 ms; paired whole-frame contrasts had both signs, and these descriptive contrasts do not establish zero overhead. Original native save/load testing failed the complete selected-state criterion in all fourteen dependent cycles, although the checked saved records matched their reference values. A separate traversal and scheduling correction passed 36 dependent cycles in twelve processes, after boundary and preserved-file diagnostics. The results distinguish component propagation, sampled rendering, workload timing and application of serialized state. They establish behavior in the declared fixtures without demonstrating physical intermediate-surface fidelity, general restoration reliability or reproduction of the historical Windows application.