Looking at the cohort definition, it's supposed to be "users who completed at least one transaction," but some rows in the cohort have zero transactions, which violates the definition. These shouldn't exist if the filter was applied correctly. Should we investigate whether there's a timing mismatch between when the cohort was formed and when the transaction data was captured?
What happened to LatAm around 2012?
Missing Brazil data after 2000
Which source is
observed coming from?Yes.