Essay
Cohort charts without vanity spikes
Retention curves lie politely when campaign weeks are allowed to redraw history. Here is a filter set that keeps Product-Led Growth Analytics honest.
Vanity spikes appear when a cohort’s week-one retention jumps because a promotion pulled in curious clickers who never intended to stay. Marketing did its job; your chart should not pretend those users were the same species as organic signups.
Separate acquisition channels in the cohort key
If you only cohort by signup week, you invite noise. Add channel or campaign family as a second split for any chart that might enter a leadership room.
Freeze the window
Decide whether day-7 means calendar days or active days, then freeze it. Quietly changing windows is how teams accidentally manufacture improvement.
Refuse decorative smoothing
Moving averages can be useful in exploration notebooks. They should not be the only line on a stakeholder chart. Show raw cohort points beside any smoothed guide.
Vector Dockcore habit
In Module 3 of Signal-to-Scale Lab, pairs submit a “spike autopsy” — one chart that looked good until exclusions were applied. The embarrassment is instructional.