An event-quality checklist before you trust cohort results

Before analysis, verify that an event means the same thing across app versions and platforms. A renamed button, duplicate mobile callback or server retry can shift a trend without any real customer change.

Check identity stitching next. Anonymous browsing and authenticated activity may belong to one person. If merging is inconsistent, users can appear new twice or seem to disappear after login.

Inspect timestamp timezone, late-arriving data, bot traffic and internal accounts. Document exclusions so another analyst can reproduce the cohort.

Finally, compare event totals with an independent operational source where possible. Analytics will never be perfectly clean, but known limitations should be visible before decisions are made.