Human-on-the-Loop (HOTL)

A model in which the system may operate without prior approval for every decision while a person observes its behavior and can intervene, correct, or stop operation. Oversight needs access to signals, context, and an override path that works within the acceptable harm window.

HOTL differs from human-in-the-loop because action does not wait for human confirmation, and from human-in-command because it supervises operation without necessarily deciding the global policy for use. A passive dashboard is not enough: alerts, thresholds, escalation, and authority to act must be defined before an incident.