Adding more communication pathways between agents doesn't necessarily improve multi-agent performance. A controlled, reproducible experiment across 50 runs reveals that recovery remained remarkably stable from 20% to 100% relationship density. However, as the network became denser, the fraction of edges actually used fell sharply—exposing a gap between configured connectivity and behavioral connectivity.

The author built a controlled experiment isolating relationship density from confounding variables using a fully deterministic agent policy instead of live model calls. Findings show that more communication pathways don't automatically mean better performance. The real engineering question isn't how many connections exist, but how many actually carry information.

The experiment uses a fixed count of eight agents and evaluates five density levels (20%, 40%, 60%, 80%, 100%) with ten trials each. All data comes from real benchmark runs on Python 3.12, CPU-only, with zero external API calls. The system uses connected Erdős–Rényi random graphs with no hidden hubs or hand-tuned rules.

This work targets those selecting multi-agent topologies by gut feeling. The complete code and pre-specified test protocol are available at the GitHub repository for exact reproduction.