Describe a situation where you had to dive deep into metrics to find the root cause of a production regression. What data did you pull, how did you validate hypotheses, and what was the outcome?