Retention Playbook

• If totalIntervals exceeds the date range, the tail cells are always zero and look like a drop. Match totalIntervals to the range.
• The most recent cohort hasn’t had its full retention window yet — note as partial rather than a regression.

1. Isolate the cohort

targetEntity / returningEntity use {type: "events", name: "<event>"} and nest in retentionFilter. Compare affected cohort(s) to prior baselines side by side.

2. Event vs. users

Is the drop in the activity event itself or in the cohort doing it? Create or reuse a cohort for the affected period (posthog:cohorts-create / -list), then run posthog:query-trends on the activity event filtered to that cohort.

3. Split the dropout

Run posthog:query-lifecycle scoped to the affected cohort to separate “never retained” (new users who didn’t return) from “lost later” (returning users who churned).