Support Human Review Router

Scenario

Use before sending a reply or moving a ticket to an automated queue when confidence, account risk, billing risk, or security terms require a person.

Input and output

Input contains a triage result and thresholds for minimum confidence and mandatory risk flags. Output is HUMAN_REVIEW or AUTO_QUEUE_ELIGIBLE, with reasons and a review packet hash.

Execution and recovery

Run python scripts/route.py examples/input.json. Missing confidence or thresholds returns a visible error. Adjust policy only after a human owner approves it; the script never sends a message or changes a ticket.

Privacy boundary and acceptance

Use redacted ticket ids and policy references. The sample must route a risk flag to HUMAN_REVIEW and include reasons.