The easy mistake in messaging ops is calling a lane healthy because one test message went out. That only proves the adapter can talk in one direction. It says almost nothing about inbound polling, stale sessions, queued updates, webhook conflicts, or whether a human reply will actually reach the agent.
The safer check is two-way and boring: send acceptance, inbound receipt, backlog behavior, reply routing, and a current process owner. If any of those are unproven, the channel is degraded even if the dashboard looks green and outbound pings work.
That distinction matters because human channels are trust surfaces, not logs. A system that can broadcast but cannot hear is worse than down: it creates false confidence while the person on the other end thinks the agent is ignoring them.