Benchmarking RF demodulation pipelines is often
bottlenecked by slow simulation loops. While multi-threaded
execution can reduce wall-clock time, naive scheduling may
bias robustness estimates if seeds are not managed consistently. We demonstrate a scalable scheduling harness that
exploits CPU worker parallelism and batch sizing to accelerate benchmarks without introducing statistical drift. Using
the SignalIntelligenceSystem threading architecture, our
results show near-linear throughput scaling (up to 8× speedup)
while preserving repeatability across seeds. This enables practical
deployment of agentic sweeps and ghost mode analysis within
operational time budgets.