Before You Simulate: A Pre-Study Benchmark for Large Language Model Stability in Political Role-Playing Simulations | Synapse