← All questions
EvaluationHard· Statistics· Experimentation· Metrics

Your target bucket improved by 15 points on n = 100, but your control bucket - queries the new component should never touch - also moved by 4 points. Do you ship?

It is critical to distinguish between signal and noise. When the control bucket shows movement, it suggests that the observed changes might be due to sampling variance. Performing a statistical check on the standard error helps prevent making decisions based on unreliable data, highlighting the need for more samples to reach a definitive conclusion.

Code playground (Python)