Human-in-the-Loop

Human-in-the-loop (HITL) checkpoints are a first-class primitive in matic, not an optional safety net layered on after the fact. When a work item reaches a checkpoint, execution pauses, a preview is surfaced to a designated human through a Channel, and work resumes only after that human responds. Each interaction produces an audit entry, and directional responses also produce a Decision record, giving operators an immutable trail of human judgment applied to agent work.

Checkpoint Placement

Checkpoint Placement explains where HITL checkpoints can be defined, including work item specs, policies, and activation gates, and how to decide which workflow points require human sign-off.

Dual-Write Rule

Dual-Write Rule covers how every HITL response writes an audit entry and, when the response is a directional choice, a Decision record linked back to the checkpoint with the hitl_checkpoint flag.

Escalation and Timeout

Escalation and Timeout describes what happens when a human does not respond, including timeout thresholds that escalate from operator to owner and eventually block the work item.

Pause-Resume Flow

Pause-Resume Flow describes the runtime mechanics of pausing a work item at a checkpoint, surfacing a preview and approval request through the notifier, and resuming or revising work after a human response.

Permission Levels

Permission Levels defines which human permission level is required for each class of HITL decision, from standard delivery approvals to baseline violation overrides and policy exceptions.

Preview Surfaces

Preview Surfaces explains how checkpoint previews are rendered and delivered through Channels so humans can review and respond without direct access to the org host.

Regression Check Overrides

Regression Check Overrides documents the protocol for overriding a baseline violation at a HITL checkpoint, including the owner-level permission requirement and the baseline_override: true decision flag carried through delivery records.