The method
How Kindred matches people.
Three stages: filter out the impossible, rank what remains by evidence, then learn from what actually happens. No compatibility scores invented from questionnaires.
The dealbreaker filter
Some things people report about themselves are reliable, and they are the ones that make or break a relationship structurally: whether you want children, do not, or already have them; the relationship structure you are looking for; your faith and whether a partner must share it; how far you are genuinely willing to be from someone; and substance use.
Kindred asks about exactly these, once, in a short questionnaire. Anyone whose answers make a pairing structurally impossible never appears in your pool, and you never appear in theirs. No time spent, no hopes raised.
Behavioural signals and reciprocity
Within your filtered pool, ranking is built from behaviour rather than self-description: signals from accounts you explicitly choose to connect, what your activity on Kindred reveals about who genuinely draws your interest, and reciprocity modelling, which estimates the probability that interest runs in both directions. An introduction only makes sense when it does.
To be plain about what this involves: this is profiling, and you deserve to know that. It is disclosed in full before you join, it runs only on data you have consented to connect, you can opt out of account connections, and a privacy impact assessment precedes launch. What we will not do is sell this data, share it with advertisers, or use it for anything other than matching and the published research you have consented to.
Learning from real outcomes
This is the stage the industry skips. After each introduction, Kindred follows up on what actually happened: did you exchange contact details, was there a first date, a second, are you still in contact at 30 and 90 days, and, eventually, at six and twelve months. Those answers, given with consent, retrain the ranking model, so the system learns from real relationship progression rather than from clicks.
Because Kindred has no engagement metrics to inflate and no advertisers to serve, there is no reason to answer these follow-ups strategically and no incentive for us to misuse them. Honest answers make the next person's match better. That is the entire loop.
What we will not claim. Published research has established that no algorithm can currently predict, before two specific people meet, whether they will want each other. Kindred does not pretend otherwise. Version one stands on the dealbreaker filter, reciprocity, and honest positioning, and the outcome loop generates the evidence for whether behavioural matching does better. Whatever we find, including nothing, gets published.
Verification and safety
Every member verifies their identity before matching begins. At launch scale, verification is reviewed by a person rather than an algorithm. In-app reporting, a documented incident-response protocol, and active moderation are in place from the first closed alpha onward, and safety tooling grows ahead of the member base.
No engagement machinery
The product principles are absolute: no infinite swiping, no paywalled visibility, no artificial scarcity, and no notifications engineered to pull you back. Introductions are curated and finite, because the goal is for you to leave.