Statistical
Peeking Control
Pre-calculate sample size and wait for significance (p < 0.05) before declaring a winner.
Metric
Power & Significance
Technical
Error Correlation
Use HogQL to verify that winning variants didn't increase technical friction or API errors.
Metric
Technical Guardrail
Monitor for Sample Ratio Mismatch. If split is not 50/50, the assignment logic is compromised.
Metric
Assignment Balance