GMT cohorts and return ghosts
Day-one retention can move when clocks and user IDs move—not when the product does.
Teams serving users across the UK often mix UTC warehouse timestamps with local product calendars. A cohort day that spans the wrong midnight shifts every retention cell without anyone shipping a feature.
SDK upgrades that change anonymous-to-known stitching create similar ghosts: returning users appear as new cohorts, and week-four rates look healthier or worse depending on merge rules.
Before celebrating a retention lift, list SDK versions, identity rules, and timezone settings beside the chart. If those moved, treat the chart as suspect until you rebuild the cohort from raw events.