Human-in-the-loop

In a human-in-the-loop system, automation does the preparation and a person makes the consequential decision. The pattern specifies where the pauses are, who is authorized to act at each one, what information they see, and what happens if no one acts. Done well, it puts the human where judgment matters and removes them from where it does not.

The pattern is only meaningful if it cannot be bypassed. A prompt-level instruction to "ask before sending" is not human-in-the-loop; an approval hold enforced by infrastructure the agent cannot reach is. Meridian enforces approval holds at the Gateway, below the agent, so no agent, builder, or prompt can skip them.

Examples in Meridian: a recruiter approves every advance or rejection, a payroll manager approves every correction, a controller releases every audit package, and an attorney approves every redline before it leaves the company.