Experiment Verification Monitoring

Experiment Verification Monitoring

Use this skill to prevent misconfigured or unhealthy experiments from producing bad evidence. It focuses on verification before launch, canaries, A/A tests, leakage and interference checks, active monitoring, and quality roadmap metrics.

Source Traceability

Primary source: Next-Level A/B Testing by Leemay Nassery. Guidance is transformed and paraphrased from Chapter 5 on experiment effectiveness, prelaunch verification, QA tooling, canaries, A/A tests, spillover effects, and active monitoring.

Related skills:

• ab-test-design-brief for planning an experiment before verification.
• experimentation-throughput-strategy for monitoring overlap conflicts.
• trustworthy-experiment-insights for statistical credibility after results.

Reference Routing

Workflow

1. Define the experiment quality risks the platform must catch.
2. Add prelaunch verification for assignment, targeting, exposure, treatment, metrics, and user experience.
3. Add early launch canaries and active monitoring for misconfiguration.
4. Run periodic A/A tests to validate infrastructure health.
5. Track quality metrics and update the experimentation playbook.
6. Define owners and escalation paths for active experiment issues.

Output Format

# Experiment Verification And Monitoring Plan

## Quality Risks
[What errors or trust failures this plan should prevent.]

## Prelaunch Checks
| Check | Method | Owner | Pass/Fail Criteria |
|-------|--------|-------|--------------------|

## Active Monitoring
- Canary:
- Dashboards:
- Alerts:
- Leakage/interference checks:

## Platform Health
- A/A test cadence:
- Quality metrics:
- Review process:

## Escalation Rules
- Pause if:
- Restart if:
- Investigate if:

Quality Bar

• Do not rely on manual QA alone when the platform has recurring setup errors.
• Do not launch experiments without verifying assignment, exposure, and metrics.
• Do not treat A/A tests as one-time setup checks; use them as periodic health checks when platform trust matters.
• Do not monitor only final results; active tests need early health signals.