Friendship Tests Usually Measure Memory. I Wanted to Measure the Friendship.

 I had two friendship-test sites open in separate tabs. One asked whether I knew someone’s birthday and zodiac sign. The other quickly produced something that could be passed to another person. In the traffic snapshot I checked, wtfti showed about 597 visits per month and 1.13 pages per visit; bfftest.fun showed roughly 215,000 visits and 4.8 pages per visit.

Those numbers do not prove why one site traveled further than the other. But they sharpened an observation that shaped KnowUsWell: a social product needs an artifact people can share. A score, a card, a label, or a link gives the interaction somewhere to go next. A questionnaire that ends with “you got 8 out of 10” is not necessarily saying anything useful about the relationship.

Most friendship tests are really memory tests. Do you know their birthday? Their favorite color? Their zodiac sign? These questions are easy to write and score because there is one correct answer. But remembering facts about someone is not the same as understanding how it feels to be around them. You can know a friend’s birthday and still repeatedly frustrate each other when making plans. You can forget their favorite movie and still understand exactly when they need company and when they need space.

I wanted to measure that second category: patterns of interaction rather than stored facts.

I reduced those patterns to four axes: Steer, Show, Storm, and Space. Steer asks who tends to set plans or create momentum. Show looks at how care gets expressed. Storm deals with what happens during conflict. Space asks how much independence or contact feels comfortable.

These are not intended as clinical dimensions, and they do not capture an entire relationship. They are product decisions: four questions about compatibility that are recognizable, distinct enough to discuss, and compact enough to turn into a result.

The test uses sixteen question cards, four for each axis. Within every axis, two items are keyed in one direction and two in the other. That balance is deliberate. If every statement rewarded agreement, someone who habitually taps “yes” could appear to have a strong preference on every axis. Mixing positively and negatively keyed items is a simple defense against acquiescence bias. It does not eliminate response bias, but it avoids building the scoring system around one predictable behavior.

Another important decision was that both people answer for themselves. One person does not guess what the other person would say, and the test does not grade how accurately one person remembers the other. Each person describes their own way of steering, showing care, handling storms, and using space. The relationship result comes from comparing those two self-descriptions.

The obvious implementation would have produced 16 individual cards and then treated every possible pairing as unique: 16 × 16, or 256 relationship combinations. I chose not to do that. Writing 256 labels would create an impressive-looking catalog, but most of the distinctions would be difficult to explain and harder to remember.

Instead, each axis is reduced to one relationship-level distinction: the two people are either aligned or complementary on that axis. Four binary comparisons become a four-bit code, which produces 16 relationship types. The compression is intentional. The result should provide enough structure to start a conversation without pretending that a quiz has discovered a mathematically exact category for two people.

Names such as “The Ringleader” are part of the same design. A code is efficient internally, but nobody naturally shares “relationship type 1010.” A named card gives the result a handle. The label is not meant to be a verdict; it is the portable part of the experience—the thing someone can recognize, send, or disagree with.

That sharing requirement also changed the technical architecture. KnowUsWell has no accounts and no backend for answers. The first person’s answers are encoded into the URL fragment, the portion after the “#”. When the second person opens the link, the comparison can happen in the browser. The answers do not need to be uploaded to a database.

Encoding is not encryption, so I do not describe the link itself as a secret vault. Anyone who has the full link may be able to inspect or pass it along. But avoiding a backend means I do not need to collect identities, maintain answer records, or ask people to create an account just to compare four interaction patterns.

I also separated the underlying mechanics from the relationship setting. The same structure can wear a “bestie” or “couple” skin, with the wording using a {them} placeholder. Adding a new skin takes less than two hours because it does not require rebuilding the scoring or comparison logic. That constraint keeps the product focused: the context can change, but the four axes remain stable.

KnowUsWell.com is my attempt to make a friendship test about negotiation, affection, conflict, and distance—not trivia. The hardest unresolved question is also the most interesting one: when a relationship tool simplifies two people into a shareable label, what makes that simplification illuminating rather than merely flattering?