Reading retention after an SDK upgrade
Day-seven return rates swung overnight. The product had not changed—identity storage had.
An SDK bump reset anonymous IDs on Android for users who had cleared cache during a forced update. Day-one cohorts looked like brand-new users, and retention appeared to collapse.
The Cohort Integrity Review rebuilt definitions from server login history and excluded the upgrade window from executive reporting. Product decisions resumed on a narrower but honest sample.
If your retention swing arrives the same week as an SDK change, treat the chart as a suspect until identity continuity is checked.