Why we’ll never put a streak on your friendships

Etkin ran six experiments on ordinarily enjoyable activities — walking, reading — with and without the participant’s output being measured. Measurement increased how much people did. It also reduced how much they enjoyed doing it, by drawing attention to output and making the activity feel like work.

The lingering effect is the part that matters. This is not a neutral motivator with an upside while it runs. After the measurement stopped, people read less than they had before, and in one experiment avoided reading altogether. The tracking left the activity in a worse state than it found it.

Now substitute "friendship" for "reading" and read that again. That is the proposition behind every streak in this category.

Clark and Mills distinguish two norms. In exchange relationships, benefits are given to incur or repay obligation, and members keep track of individual inputs. In communal relationships, benefits follow from concern for the other person’s welfare.

Tracking inputs is diagnostic of the first kind. And where closeness is wanted, prompt repayment of a benefit actually reduces attraction — the ledger behaviour itself signals the wrong relationship type.

This is the theoretical name for why personal CRMs feel unpleasant to a lot of people. They import pipeline machinery — last contacted, follow-up due, cadence — into relationships governed by a norm where that machinery is a violation. It is not a UI problem and a nicer design does not fix it.

The distinction that survives all this is between needs and balances. "Her mum is having surgery in March" is a need — communal, and the kind of thing a friend holds. "You haven’t messaged her in 47 days" is a balance, and it is exchange machinery wearing a friendly tone.

So Kept stores the first and never displays the second. Signals like recency and frequency do exist internally, because they are genuinely useful for deciding what to bring back first — the contact-management literature has known since 2002 that these are the criteria people actually use to judge importance. They are used for ranking and never surfaced as a score.

There is also no streak on any person, no relationship-health percentage and no leaderboard. Those are the highest-leverage retention mechanics available to us and we are not going to ship them.

Kept does have one streak, and it is worth saying where the line sits. It is weekly and it belongs to the journal as a whole: one note in a week keeps it, about anyone, and after a quiet week it reads as eleven of the last twelve rather than falling to zero. It counts the habit of writing things down. It does not count a person, and nothing in the app tells you how long it has been since you spoke to someone as a number to keep up.

Etkin’s experiments are about walking and reading, not relationships. The extrapolation to friendship is ours, and it is an extrapolation.

The rewards literature is also not unanimous. A meta-analysis of 128 experiments found tangible engagement-contingent rewards reliably decreased intrinsic motivation, r ≈ −0.24; a competing analysis put the effect far smaller. The direction of risk is consistent even where the magnitude is disputed.

Given that, the choice is between a retention mechanic with a plausible chance of degrading the thing the product exists to support, and no retention mechanic. We would rather lose the engagement.