Retention Cohort Diagnostics

Retention charts move for many reasons that have nothing to do with product quality: SDK upgrades, timezone mishandling, silent user-id changes, or a new onboarding that simply renames day-one events.

This study rebuilds cohort definitions from raw event history, checks identity continuity, and tests early-session behaviours against return rates. We are explicit about sample limits and what the data cannot prove.

You leave with a shortlist of behaviours worth protecting in the product, a list of measurement artefacts to stop treating as insights, and a reusable cohort notebook outline your analysts can maintain.