BreezThru keeps humans in the loop wherever judgement matters. Two mechanisms do this: approvals and spot-checks.
An approval is a human decision on an L2 step's output. Approving lets the run continue; rejecting sends it back.
A spot-check is a sample of an L1 (autonomous) step's output, routed back for after-the-fact review. It's how the platform verifies autonomy without blocking every output on a human.
Spot-checks can be approved or rejected directly in the queue.
Every decision, whether approve, reject, override, or spot-check, is append-only and auditable. You can see the trail on the target's page and query it from Search.