Pre-Production Assessment Gates

Before any AI system enters production governance, it must pass a pre-production assessment in the GOVERN Sandbox. This gate ensures the system is properly configured and meets your organization’s governance baseline.

The Pre-Production Gate

The pre-production gate is a structured assessment sequence that every new AI system must complete before production onboarding:

1. Initial assessment (automated)
2. Policy compatibility check
3. Red team baseline
4. Benchmark baseline
5. Human review
6. Production approval

Step 1: Initial Assessment

# Register AI system in sandbox
govern sandbox add-system $SANDBOX_ID \
  --name "new-system-v1" \
  --type api \
  --endpoint $SYSTEM_ENDPOINT

# Run initial automated assessment
govern assess \
  --sandbox $SANDBOX_ID \
  --system new-system-v1 \
  --output initial-assessment.json

The initial assessment scores the system across:

Dimension	Pass Threshold
Security	≥ 70
Bias	≥ 75
Accuracy	≥ 65
Transparency	≥ 60
Alignment	≥ 70
Overall	≥ 70

A system scoring below the threshold must be remediated before proceeding.

Step 2: Policy Compatibility Check

# Test system against your production policies
govern policy test \
  --sandbox $SANDBOX_ID \
  --system new-system-v1 \
  --policies all-production \
  --test-cases 500 \
  --output policy-compat.json

This verifies:

• Each policy triggers correctly on violation inputs
• No critical false positives on compliant inputs
• Policy overhead is within latency limits

Step 3: Red Team Baseline

govern red-team standard \
  --sandbox $SANDBOX_ID \
  --system new-system-v1 \
  --output red-team-baseline.json

Minimum passing criteria:

• Injection block rate ≥ 97%
• Jailbreak block rate ≥ 95%
• No critical data extraction succeeds

Step 4: Benchmark Baseline

govern benchmark run \
  --sandbox $SANDBOX_ID \
  --suite standard \
  --system new-system-v1 \
  --output benchmark-baseline.json

Baseline results are stored and used for future regression detection.

Step 5: Human Review

The pre-production report is generated and routed to the designated AI system owner and governance team for human review:

govern sandbox report \
  --sandbox $SANDBOX_ID \
  --system new-system-v1 \
  --type pre-production \
  --format pdf \
  --output pre-production-report.pdf

The report is submitted to the approval workflow. Approvers receive an email with the report attached and a link to the approval interface.

Step 6: Production Approval

After human review, the approver signs off in the GOVERN dashboard:

Admin > AI Systems > Pending Approval > [system name] > Approve for Production

Approval is logged with the approver’s identity, timestamp, and a reference to the pre-production report.

Production Onboarding

Once approved:

# Promote system from sandbox to production
govern system promote \
  --from-sandbox $SANDBOX_ID \
  --system new-system-v1 \
  --to production

This creates the production system record with the pre-production assessment as the baseline for ongoing monitoring.