What is the standard approach to define agent capacity in the routing configuration for high-volume load tests? The JMeter script sends 500 concurrent WebSocket connections, but the system reports zero available agents even though the YAML defines capacity as 100 per agent.
routing:
predictive:
strategy: forecast
agent_capacity:
default: 100
max_concurrent: 5
The logs show capacity_exceeded errors immediately after connection. Is the capacity value interpreted differently in the API versus the UI?