The Kin Matching Algorithm: Objective Compatibility Scoring for Family Formation

Kin is our most unusual product. It exists at the intersection of some of the most personal, consequential, and biologically grounded decisions a person can make: whether to have children, with whom, through what means, and on what timeline. Most matchmaking products treat these questions as secondary to attraction and chemistry. We treat them as primary.

This post describes the algorithm behind Kin's compatibility scores — why we built it, how it works, and where we know it falls short.

Why an algorithm at all?

The case against algorithmic matching is familiar: human connection is ineffable, compatibility is dynamic, reducing people to data points misses everything that matters. We take this seriously. But we also think the case has been overstated, particularly for users with a specific goal: finding a partner with whom to build a family.

Users who come to Kin have already made a series of clarifying decisions. They want children. They have a timeline. They have views on how they want to raise children, what genetic considerations matter to them, and what life structure they're seeking. These are not ineffable. They can be expressed, compared, and scored. And when the stakes are as high as family formation, users deserve more than a stack of photos sorted by recency.

Our algorithm doesn't tell people who to love. It tells them who is worth having a conversation with.

Architecture: hard filters + weighted scoring

Kin's matching engine has two stages: hard filters and soft scoring.

Hard filters

These are binary eliminators. A pair doesn't generate a match if any hard filter fails:

• Sex preference alignment. Each user specifies who they're open to partnering with. Both directions must be satisfied — A must be open to B's sex, and B to A's.
• Seeking overlap. Kin supports multiple family formation modes: marriage, co-parenting, sperm/egg donation, adoption. At least one mode must appear in both users' seeking lists.
• Age range compatibility. Each user specifies the age range they're seeking. Both users must fall within the other's stated range.

Hard filters are not scored — they are gates. Passing all three generates a candidate pair that then enters the scoring stage.

Weighted scoring

We score seven dimensions, each weighted by its contribution to long-term family formation success as supported by literature in relationship psychology, behavioral genetics, and family systems research:

Dimension	Weight	What we're measuring
Family formation intent	25%	Alignment on marriage vs. co-parenting vs. donation; seriousness of intent
Core values	20%	Overlap across religion, political orientation, lifestyle, parenting philosophy
Children preferences	20%	Number of children desired; openness to adoption; existing children
Timeline	15%	Distance between target start dates in months
Genetic & health profile	10%	Presence of heritable conditions; disclosure of carrier status
General health	5%	Self-reported health status and lifestyle factors
Location	5%	Geographic proximity or explicit openness to relocation

Each dimension produces a score between 0 and 1. The weighted sum produces a composite score between 0 and 100. We display this to users as a match percentage.

Dimension design decisions

Why intent is weighted most heavily

The research literature on relationship dissolution consistently identifies misalignment on "want children / don't want children" as one of the leading causes of long-term relationship failure — second only to fundamental value incompatibility in studies of divorce predictors (Gottman, 2011; Dew et al., 2012). In a product specifically designed for family formation, this alignment deserves the highest signal weight. A couple who agree deeply on how they want to raise children but are in different geographies can solve the geography problem. The inverse is much harder.

Values at 20%

We use a structured values survey covering religion, political views, lifestyle (diet, substances, social life), and parenting philosophy. Values scoring is not about agreeing on every dimension — it's about not having fatal incompatibilities on the dimensions that research shows predict conflict. We score similarity, not identity.

The genetics dimension

This is the most sensitive dimension we score. Our position: users who are carriers of heritable conditions deserve to know whether their potential partner has disclosed similar carrier status, so they can make informed decisions. We don't make this determination for them — we surface the information and let them decide what weight to give it. The scoring function gives partial credit for transparency (disclosing unknown status) even when exact genetic compatibility can't be determined.

We do not use genetic data to rank-order candidates in ways that could constitute eugenic selection. The genetics dimension affects match ranking marginally (10%), and users can opt out of including it entirely.

What the algorithm doesn't do

The algorithm doesn't predict love. It doesn't predict whether two people will find each other attractive. It doesn't account for the thousands of variables that emerge in an actual relationship. A 90% match score is not a guarantee — it's an invitation to explore whether the structured compatibility translates into real connection.

We also don't use photos as a matching signal. Kin is not optimized for physical attraction — it's optimized for family formation readiness and compatibility. Users can share photos after a match is made, but they're not an input to our scoring.

What we're working on

The current algorithm is purely self-reported-data-based. We're exploring several extensions:

• Feedback loops. Incorporating explicit user feedback ("this match felt right / wrong") to tune dimension weights over time using preference learning.
• Timeline dynamics. The current timeline score is static. We want to model it as a moving target — accounting for how a user's urgency changes over time on the platform.
• Verified credentials. For medical and genetic dimensions, we're evaluating integrations with verified health record sources to replace self-report with structured clinical data.
• Fairness auditing. Ensuring the scoring function doesn't produce systematically different outcomes across demographic groups in ways that perpetuate historical inequities in family formation access.

Family formation is one of the most important decisions in a human life. We're building the tools to make it more intentional — not more algorithmic.