Journal · January 2026

Reading retention without vanity

Retention is not a mood. It is a map, and maps have legends. Throw the legend away and you can make any coastline look inhabited.

Unbounded retention asks: of the people who did a thing on day zero, how many had done it again by day N, at any point in between. It is a cumulative kindness. N-day retention asks whether they did it on day N itself. It is a stricter census. Rolling retention, depending on the vendor, is a third dialect. Most slide titles say only “retention.”

In the rooms we make students label the map in the first sentence of the readout. “N-day retention of users who completed a lesson, window frozen at 1, 7, and 28.” If that sentence feels long, it is still shorter than the argument you will have in March when someone pastes an unbounded curve next to last quarter’s n-day number.

Vanity enters when the population is quietly narrowed. Drop the users who never finished onboarding and your day-7 line lifts. That can be the right population for a quality question. It is the wrong population for a growth claim. App Analytics does not know your intention; it will happily compute either. The intention has to live in the paragraph.

Another vanity is comparing cohorts of different ages as if they were the same weather. A January cohort that has had a year to return will look more “loyal” than a November cohort that has had six weeks. Retention Cartography, when we run it as a short room, spends two evenings on nothing but age alignment.

We are not purists about which map is morally better. We are purists about not switching maps mid-sentence. If your board pack did that last quarter, bring the pack to Cohort Instrumentation. The hearing is unkind in a useful way.

Return to the journal.