Constraint-Aware Diffusion for Valid Graph Trajectory Generation

This framework combines conditional diffusion models with a symbolic layer to ensure generative graph trajectories remain structurally feasible. In compact graph testing, the model reduced invalid probability mass to 0.002996 through hard filtering and projection-based repair.