Most dating apps decide who you see. Ours sorts profiles by four rules you can read, the same for everyone, and stops there. That is a deliberate choice and it costs us something, so it is worth explaining rather than listing as a feature.
What does a ranking algorithm actually optimise?
A recommendation system needs a target. On a dating app the available targets are engagement metrics: time in app, messages sent, swipes per session, days retained. None of those is "you met someone good". They are proxies, chosen because they can be measured every day, and a system trained on a proxy will eventually optimise the proxy at the expense of the thing it stood for.
This is not a conspiracy theory about dating apps. It is arithmetic. If people who find a partner stop opening the app, and the system is rewarded for people opening the app, the system learns something about people who find partners. Nobody has to intend that for it to happen.
The second problem is specific to us. A ranking model learns from behaviour, and behaviour towards trans people on general dating apps is not neutral. Feed it engagement data and it will faithfully reproduce whatever hierarchy that data contains: which profiles get opened, which get skipped, whose photos get lingered on and why. On a platform made for trans and non-binary people, an engagement-optimised feed would rebuild the exact ranking people came here to get away from, and it would do it invisibly, with the authority of a machine.
What do we do instead?
Four plain rules, and no score. Inside the radius you set, profiles with a photo come first, then profiles that answer a profile question, then, if nobody in your results is within 100 km, people who say on their profile that they are open to talking at a distance, and within each group the closest person comes first. The reason is practical: a like is given to a photo or to an answer, so a card with neither gives nobody anything to respond to. The third rule follows a setting each person chooses on their own profile. There is no hidden score, no boost, no "highly compatible" badge and nothing you can buy to move up the list.
You can filter (age, whether there is a photo, whether the account is verified) because a filter is something you chose and can see. That is the difference we care about. A filter is legible. A ranking is not.
We also show the real number of people in your radius, unpadded. If there are eleven people within fifty kilometres, it says eleven. Every dating app knows the number that makes a city look alive, and every dating app is tempted to show that one instead. Ours says eleven, and you decide whether eleven is worth your time. See how it works for the rest of the mechanics.
What does this cost?
Being honest about the downsides is the only reason a post like this is worth reading.
Sorting by distance is weak in dense cities. In a large city with hundreds of accounts nearby, "closest first" inside each group is close to arbitrary. Somebody four hundred metres away is not more relevant than somebody two kilometres away, and past a certain density the grid becomes a wall of people with no help navigating it. We have no clever answer to this yet. Filters and profile answers are what we offer, and they are not a complete substitute.
New accounts get no boost. Many apps push a new profile to the top for a day or two, which genuinely helps someone get started. We do not, because that mechanism is the same one used to sell visibility later. The cost is real: joining is slower here.
We cannot show you "people like the ones you liked". That feature is often good, and we are giving it up. We would need to build a behavioural model of you to provide it, and we would rather not hold that model at all than promise to use it carefully.
Engagement is probably lower. An app that does not try to keep you scrolling does not keep you scrolling. We think that is the correct outcome and we are aware it is also the kind of thing a company says when its numbers are small.
The part that is not about algorithms
There is a version of this argument that is mostly branding: no algorithm, no ads, no dark patterns, we are the good ones. That version is cheap, because nothing in it can be checked.
What can be checked: the app shows the unpadded count, the sort follows the four rules above, the filters are the ones you can see, and profile answers are visible according to rules you set yourself on the visibility screen. What we keep and what we do not is written on the privacy page, including the things that would be convenient for us to keep and that we deliberately do not. Moderation numbers, including the ones we would rather not publish, are on the transparency page and are read from the database rather than typed in.
Those are the claims to hold us to. Not the absence of an algorithm, but the presence of things you can look at.
Wren · Editor
Wren is the name the T4TSpace blog is published under. The posts are written by the small team that builds the app, and signed with one name so the voice stays the same from one piece to the next.