Temporal Work Region Fabric: Persistent Spatial Work Objects for Incremental Virtual-GPU Execution

TWRF studies a virtual-GPU organization in which temporally reusable spatial computations are represented as persistent Temporal Work Regions. A TWR retains identity, spatial extent, state/output, versioned inputs, dependencies, and execution state across frames. The final software line passes 35/35 registered tests. The 14-case raster matrix achieves 14/14 execution-set parity, 14/14 output parity, zero dependency-audit failures, and 14/14 full-recompute oracle parity. Under the declared timing model, TWRF has lower derived cost than executable software incremental Baseline C in all 14 cases, while temporal caching remains lower-cost throughout the default matrix and full recomputation remains lower-cost for every non-static case. A 135-setting sensitivity campaign evaluates region size, control-plane cost, and State Store latency. These are derived architectural results, not physical-GPU measurements.