12 January 2026 · Retention

Retention curves that flatter the launch week

A smooth D30 line after a feature launch is often a souvenir of the listing, not proof that residents changed their week.

When a United Kingdom consumer app ships a visible feature, two populations arrive. Residents already had a reason to open you on a Tuesday. Tourists arrived because the store put you on a shelf. App Analytics tools, left on defaults, pour both into one cohort and then congratulate the product team for “retention” that will not repeat in March.

Broker Kernelhub teaches a split that sounds bureaucratic and saves careers. Label the first-open source. Keep listing-driven opens in their own family of curves. Do not average them with residents unless you are writing fiction for a board pack.

The launch-week bulge

Launch week inflates D1 because curiosity is cheap. D7 still looks sturdy if the feature has a one-time setup. D30 collapses when the setup is finished and the listing has moved on. If your chart uses unbounded retention, the bulge is even kinder: anyone who returned once in thirty days counts as kept, including people who opened by accident after an OS update.

We ask students to publish two D30s or none. The resident curve is allowed to be modest. Modest and true beats a blended line that cannot be forecast.

What not to do

Do not delete tourists from history. They paid in attention. Report them. Just stop letting them vote in the metric you use to staff a squad. The Signal Ledger has a slot for one retention family; pick the residents unless you are a growth team whose job is listing tourism.

Return to the journal or sit Cohort Signal Studio if you want the window exercises with a tutor interrupting you.