How desirable are you, statistically?

Your answers go through a model that learned from real matchmaking pairs, meaning actual people a professional matchmaker put together. It runs on your own device rather than on a server.

Each answer has its own smooth curve, so moving one step never causes a sudden jump. How much each question counts comes from research into what people actually choose in a partner, the same weighting the "what could I get" page uses: looks and warmth carry the most for women, money, drive, height and social ease the most for men, and strong traits together count extra. Your result is then placed against 214,733 real dating profiles (35,626 men and 179,107 women) to work out what share of people you rank above.

If you skip a question, you are scored as if you sit in the middle of the pool on it, and the estimate simply gets less certain. Skipping never counts against you. Questions about taste, like politics, faith or wanting kids, do not move the score either way: they decide who fits you, not how sought-after you are.

Each trait keeps the shape the model learned for it, with three rules applied on top.

First, each one can only run in one direction: richer, warmer and better looking always make someone harder to get, never easier. Left to itself the raw model does sometimes say a very rich woman is easier to get than a comfortable one. That happens because almost nobody in its training data was that rich, so it has nothing to go on there. It is a hole in the data rather than something people actually think.

Second, how much each trait counts for is set from research into what people actually choose rather than from the model, because for some traits the model genuinely cannot tell. The women’s money curve in the training data is completely flat from “stable” upward. In that research, a woman’s standing rides mostly on looks and warmth, a man’s mostly on money, drive, height and social ease, and traits like openness barely move anyone’s.

Third, great traits stack at a premium. Someone gorgeous but broke, or rich but plain, is far easier to find than someone who is both, so each strong trait costs more when the rest of the build is already strong. Maxing one dial is cheap on an ordinary person and expensive on an excellent one.

Age is the one thing not taken from the model at all. The model thinks women are most in demand at 30. That is simply the most common age of the people in its training data showing through, rather than what anyone prefers. So age uses a separate curve of how sought-after each age really is, plus an extra cost for a big gap from your own age.

Politics, faith, kids, and how tall a woman is, are treated as personal taste rather than as better or worse. They cost you nothing at all. What they change is how many people match what you asked for, which is the line underneath those rows.

Roughly speaking, models like this explain about a quarter of what actually happens and miss the other three quarters. The trait curves come from real matchmaking pairs, the weights come from published research into what people choose, and the blend is judgment. It puts people in broadly the right order more often than not, gets plenty of individuals wrong, and is a long way from a verdict on anyone.

It also isn’t predicting marriages it watched happen. It predicts a score the matchmaking data implies. Two real-world adjustments sit on top: past the mid-50s your score is scaled down by how few people realistically date your age, and a woman taller than 6′0″ is scaled by how few men are taller still. Non-binary results are the average of the male and female versions, which is a stopgap rather than something the data measured, so treat that number as looser still.

Then a model that has never met you put eighteen self-reported answers into an order, which is all that happened.

It is worth looking at what it had to work with. It does not know whether you are funny, whether you are kind when being kind costs you something, how you handle a bad week, what you are like at 2am, or who has already loved you and why. People do not choose partners by ranking them, and most of what actually decides it is missing from these eighteen questions.

The number is measured against a reference pool of people currently on dating apps. Being ranked low there says something about one market at one moment, on a handful of attributes. Plenty of people this model would rank low are married to someone who thinks they got the better end of it.

If this has landed harder than a website ought to, close it. That is a fair sign it is measuring the wrong thing for you. And if you are going through something heavier than a calculator can account for, talk to someone you trust, or a professional.

No. Everything runs in your browser. The model, the scoring and the pool comparison all happen on your own device, and nothing you enter is transmitted or stored anywhere. Nothing is saved between visits either, which is why a refresh starts you over. A share link only carries what you chose to put in it, and only travels if you send it. The portraits are AI-generated composites, so none of them are real people.

This is an indicator rather than a prediction, and it is for interest rather than for decisions. Nothing here is advice of any kind, medical, psychological or otherwise, and no result forecasts how anything will actually go. The model learns from one matchmaker's clients and ranks you against people currently on dating apps, which is a particular slice of the world and not a census. For adults.