spatial-signal-learnability-simulation-study: Simulation code for 'When Spatial Signal Is Not Enough: Learning Validity, Model Advantage, and Computational Feasibility in a Staged Simulation Study'

Simulation code for a staged study that distinguishes predictive spatial opportunity, DCRNN learnability, realized comparative advantage over a negative-binomial GLM, and operational feasibility in a 49-node healthcare-utilization simulation. Implements four evidence stages: oracle spatial diagnostic (4 seeds), Gate A 40-replication comparison, Phase R2 continuous-training extension (400 epochs), and a DD feasibility pilot. All stages are covered by a frozen protocol with seven amendments and a decision record documenting non-launch of the confirmatory analysis.