Human-after-the-Loop (HATL)

What is HATL?

Human-after-the-Loop (HATL) is a governance pattern where agent actions are reviewed after execution. Unlike HITL, it doesn't block execution, making it ideal for quality audits and compliance reviews.

When to Use HATL

• Quality Assurance - Review output quality
• Compliance Audits - Verify regulatory compliance
• Training Data - Collect labeled examples
• Performance Monitoring - Track agent behavior
• Low-Risk Actions - Where delay isn't acceptable

Enabling HATL

Via SDK

from turingpulse_sdk import instrument, GovernanceDirective

@instrument(
    name="customer-support",
    governance=GovernanceDirective(
        hatl=True,
        reviewers=["qa@company.com"],
    )
)
def handle_query(query: str):
    return agent.respond(query)

Via UI

1. Navigate to Governance → Policies
2. Click Create Policy
3. Select Human-after-the-Loop (HATL)
4. Configure workflow, reviewers, and sample rate
5. Save and enable

Sample Rate & Conditions

To review only a percentage of executions or apply conditional triggers, configure policies in the platform:

1. Navigate to Governance → Policies
2. Create a HATL policy for your workflow
3. Set sample rate (e.g., review 10% of runs)
4. Add conditions (e.g., always review low-confidence outputs)

The SDK enables HATL on a workflow. Sampling and condition logic is managed in the platform, giving you flexibility without code changes.

HATL Workflow

1. Execution - Agent runs normally
2. Queue - Selected runs added to review queue
3. Review - Reviewer examines input/output
4. Action - Acknowledge, Flag, or Escalate

Review Actions

HATL Configuration

Next Steps

• Human-in-the-Loop (HITL)
• Human-on-the-Loop (HOTL)
• Review Queue