Delegation Masking: Why Your LangChain Callbacks Lie About Sub-Agent Failures

When one LangChain agent delegates to another, the parent callback may only see the wrapper return value and miss failures inside the sub-agent.

A sub-agent can fail or return empty output while the parent still logs success, creating a monitoring blind spot.

The fix is to validate outputs at the delegation boundary and link traces across agents with a shared correlation ID.