My 14-agent system's results:
- OrchestratorAgent scored 4/100 — single point of failure with 3 collusion cliques
- FaceDetectorAgent scored 44/100 — high cascade depth
- EvolutionAgent scored 50/100 — 100% blast radius
Stop guessing which agent to fix. Let the data tell you.
5 features in 5 days. 84 tests. CrewAI + LangGraph.
github.com/surajkumar811/swarm-test
AI #MultiAgent #Reliability #OpenSource
Top comments (0)