How matching actually works
5 min read
People assume the score counts how many of your ten words overlap. It does not. We read a portrait out of your words, then compare two portraits for semantic overlap.
First, choose what you are looking for
Friendship, dating, activity partner — the three are never mixed in one list. Someone who wants a hiking buddy and someone looking for a relationship only exhaust each other.
Gender is filtered both ways: you choose 'women', and the other person must also be open to being seen by men before either of you appears in the other's list.
Second, ten words become a portrait
The model returns three things: an 80-character description, three concrete traits, and one question you could ask someone else.
It reads the relation between words, not the sentiment of each one. 'Late night' with 'code' and 'quiet' produces a different person from 'late night' with 'beer' and 'friends'.
Third, compare two portraits into 0–100%
We compare the two portraits semantically and rank by the resulting resonance. The same pair of users always sees the same number — it does not change on refresh.
Scores cluster between 55 and 95: the highest is not 'most similar', it is 'overlapping where it matters, different elsewhere'.
What we never show the other person
Your ten words never appear on anyone else's page. They see your name, gender, portrait and one 'need'. You do not see their words either.
Email is never public. You can turn discovery off, block or report at any time; blocked people stop appearing in your list.
Questions
- Why can't I see shared words?
- Deliberate. A shared-word list makes people stare at literal overlap instead of trying to understand a person.
- Why is my score with someone unchanged?
- It depends only on the two portraits. It changes when one of you rewrites their ten words.
- Does a high score mean we will get along?
- No. It means your portraits are close. The rest depends on the first message.