Is this a personality test?

No. A personality test describes one person. This measures four dimensions for each partner separately and then compares the two sets of answers, because the thing being assessed is the relationship rather than either individual. Joel, Eastwick and colleagues (2020) tested that distinction directly across 43 longitudinal datasets: relationship-specific measures accounted for up to 45% of the variance in relationship quality, a person's own individual traits for 21%, and a partner's individual traits for roughly 5%. Once relationship-specific answers were available, individual traits added no further predictive power.

Does it predict whether a relationship will last?

No, and it is built not to. It identifies behavioural risk factors and structural frictions — patterns that research associates with difficulty — and reports where two people fit easily and where they will have to work. It produces no probability, no forecast and no verdict. The same body of research is the reason: the 2020 PNAS analysis found that change in relationship quality over the course of a study was largely unpredictable from self-report, so any instrument claiming to forecast a trajectory is claiming more than the evidence supports.

What is the compatibility score out of 100?

It is an index describing fit across four dimensions, not a percentage likelihood of anything. A score of 72 does not mean a 72% chance of staying together, and it is not a grade. It is a position: a summary of where two people's answers align without effort and where they diverge in ways that tend to cost something. The detailed report is more useful than the number, because it names which specific frictions produced it.

How is this different from Myers-Briggs or the Enneagram?

Those assign one person a type, and types are cut from continuous distributions at the midpoint — where most people actually sit — so small answer changes reassign someone. The deeper difference is the unit of analysis. A type describes an individual, and compatibility is a property of a pair; two accurate descriptions of two people do not combine into a description of their relationship. No published evidence base supports type-pair compatibility claims for either framework.

What about the five love languages?

Impett, Park and Muise (2024) reviewed the evidence against Chapman's three core claims — that each person has one primary language, that there are five, and that matching them improves a relationship — and found support for none of them. The matching claim fails most clearly: couples whose primary languages agreed did not report higher satisfaction than couples whose differed. What does hold up is simpler — receiving expressions of affection predicts satisfaction, in whatever form it arrives. This assessment computes no language-match score.

Is my partner avoidant?

Probably a less useful question than it feels like. The internet sorts people into fixed Anxious and Avoidant types; the research measures attachment anxiety and avoidance as dimensions a person moves along, which shift over time and differ between that person's relationships. What gets called avoidance during a specific argument is usually describable as something more concrete: withdrawal under physiological flooding, where sustained arousal narrows access to language, or a breakdown in positive dyadic coping where bids for connection stopped being turned toward. Both are measurable and both change, which a fixed label cannot express. Somatic Regulation and Relational Communion measure them separately.

Isn't communication the key to a good relationship?

Communication is necessary and it is not sufficient. Where a conflict is driven by escalation, unchecked assumptions, or bids for connection being missed rather than refused, talking differently genuinely resolves it, and that covers a great many arguments. It stops working at the structural boundary. Schwartz showed that values sit on a continuum where opposing positions are in genuine motivational conflict, such as Conservation against Openness to Change. Two people on opposite poles are not misunderstanding each other; often they understand each other exactly. No conversational technique moves the midpoint to a place that satisfies both, because no such place exists. The useful step is knowing which of the two kinds of friction you are in.

Can it tell me when my relationship will fail?

No, and it is built not to. The assessment evaluates statistical trajectories of structural friction — whether the measured gaps between two partners point toward friction that recurs or toward a pattern that holds — and reports one of three directional bands. It names no date, no breaking point and no probability. Joel, Eastwick and colleagues (2020) found change in relationship quality largely unpredictable from self-report across 43 longitudinal datasets, so a questionnaire naming a month would be claiming more than the largest study of its kind could deliver. What the report does say is which specific gap puts a couple where they are, which is checkable.

Has this particular instrument been validated?

Not yet, and no accuracy figure for it is published. The research it is built on is peer-reviewed; this specific implementation has not been tested against outcomes. Validation work is underway: predictions are frozen in an immutable snapshot when a report is generated, and respondents are later asked about real outcomes, so each prediction can be scored against what the model said at the time rather than a later version of it. No figure will be published until the sample supports one.

What happens to the answers?

Each partner's individual answers are never shown to the other. Only the four dimension scores and the comparison built from them are shared between two people in a couple's report. The privacy notice describes what is stored, for how long, and what is shared with any third party.

How long does it take?

About four minutes. Each partner answers 37 statements alone, on their own device, without seeing the other's responses. The comparison needs both sets, so the report appears once the second person has finished.

Back

The psychology behind the test

Most compatibility quizzes ask what you like. This one asks what you and your partner do under pressure, and compares the two answers, because the research says that is where the information is.

The bottom line

Relationship-level measures predict relationship quality far better than personality does. The unit of analysis is the pair, not the person.
Each partner answers 37 statements alone. Four dimensions are scored for each person, then compared — and the comparison works differently on each dimension.
Arguments that escalate map to Somatic Regulation; feeling like roommates maps to Relational Communion; friction over money, children or routine maps to Structural Fidelity and Values; feeling steamrolled maps to Relational Agency.
The output is a set of behavioural risk factors and structural frictions, not a verdict. No probability, no forecast, no prediction of whether a relationship lasts.
The instrument itself is not yet validated. No accuracy figure is published, and the section on validation explains what would have to be true before one is.

The problem it was built for

People rarely look for something like this out of curiosity. They look when something specific is already wrong and they do not yet have words for it.

The same argument, endlessly
Different subjects, identical shape. It is rarely about the dishes, the message, or the money. A recurring argument usually means two nervous systems responding to threat in ways that make each other worse.
The anxious and avoidant loop
One partner moves toward conflict to resolve it; the other moves away to survive it. Pursuit reads as attack, withdrawal reads as abandonment, and each response is the thing the other cannot tolerate. Both people are behaving reasonably. The combination is not.
Shutting down, stonewalling, going blank
Not indifference. Physiological flooding: heart rate climbing past the point where language and perspective remain available. The person who goes quiet is often the most overwhelmed in the room.
Dating app fatigue
Matching constantly, connecting rarely. After enough rounds the suspicion sets in that the problem is a pattern being repeated rather than a run of bad luck.
Good on paper, hard in practice
Shared values, real affection, and a daily friction neither person can name. Usually a mismatch on a dimension nobody thought to look at.
Deciding whether to stay
Trying to tell a fixable pattern from a structural incompatibility, before committing further to either.

Empirical foundations

Four bodies of peer-reviewed work decide the shape of this instrument: one says to measure the pair rather than the people, one says to start with the body, one says how two positions should be compared, and one says why values behave differently from everything else.

Relationship-level measures outpredict personality, by a wide margin

Joel, Eastwick et al. (2020). Proceedings of the National Academy of Sciences, 117(32), 19061–19071.

Dyadic, relationship-specific measures: up to 45%
A person's own individual traits: 21%
A partner's individual traits: ~5%

What they did: Random forest models run over 43 longitudinal datasets drawn from 29 laboratories, covering 11,196 couples. It is the largest coordinated attempt made to find out which self-report measures actually predict relationship quality, and it was designed to settle the question the figures above answer: whether what matters is who the people are or what happens between them.
What they found: Relationship-specific measures — perceived partner commitment, appreciation, sexual satisfaction, perceived partner satisfaction, conflict — accounted for up to 45% of the variance in relationship quality measured at the same time. Measures of the individual accounted for 21%, and a partner's individual traits for roughly 5%. The strongest result is not any of those numbers: once a person's relationship-specific answers were available, individual traits added no further predictive power at all.
What it changes here: The unit of analysis is the pair, not the person. Each partner answers alone, but every score that matters is a comparison between two sets of answers rather than a description of one. It is also why this is not sold as a personality test, and why a single person's result is the least interesting thing the system produces.

Worth knowing. The 45% figure is for measures taken at the same time as the outcome, so part of it is two readings from one person on one day — it is not a prediction of the future. Measured against relationship quality at the END of a study, the same variables accounted for up to 18%, and individual traits for 12%. The same paper found that change in relationship quality over the course of a study was largely unpredictable from self-report. That finding is why nothing here forecasts a trajectory, and why no accuracy percentage from this dataset appears anywhere on this page.

Conflict physiology carries information that satisfaction scores miss

Levenson & Gottman (1983, 1985); Gottman & Levenson (2000).

What they did: Couples were brought into a laboratory, wired for heart rate, skin conductance and gross motor activity, asked to discuss a real and unresolved disagreement, and then followed for years — in the longest study, fourteen. The measurements were taken while the argument was happening rather than reconstructed from how either partner described it afterwards.
What they found: Physiological arousal during the conversation carried information that self-reported satisfaction did not. Couples whose bodies escalated most during conflict tended to decline in satisfaction over the following years, even when they described themselves as satisfied at the time. Sustained heart rate above roughly 100 beats per minute is the conventional marker for diffuse physiological arousal, the state usually called flooding, in which access to language and perspective-taking narrows. The associated behaviours — criticism, contempt, defensiveness and stonewalling — are the four horsemen, and stonewalling is the visible signature of a flooded nervous system.
What it changes here: Somatic Regulation is the first dimension rather than an afterthought, and it asks about the body during an argument rather than about communication skill in the abstract. It is also the dimension that most often explains a couple who agree on everything and still cannot finish a conversation.

Worth knowing. The headline accuracy percentages from this line of work have been criticised for being fitted to the same couples they were derived from, which inflates them. The finding that conflict physiology matters has replicated well; the specific hit rates have not. This page therefore cites the finding and not a single accuracy figure from it.

Interpersonal behaviour has two axes, and they fit together differently

Bakan (1966); Carson (1969). The interpersonal circumplex.

What they did: Decades of work arranging interpersonal behaviour on a circle defined by two independent axes: agency, running from dominance to deference, and communion, running from warmth to hostility. Nearly every interpersonal trait measure developed since can be located somewhere on that circle, which is what makes it a structure rather than another list.
What they found: The two axes behave differently when two people meet, and the difference is the whole point. On agency, complementarity runs by opposition: one person taking the lead invites the other to follow, and two people competing for the same position collide. On communion, complementarity runs by correspondence: warmth invites warmth, and coldness invites coldness. A single similarity score cannot express both at once.
What it changes here: Two of the four dimensions are compared in the way the theory says they behave, which means a difference between two people can be an asset on one dimension and a cost on another. This is the published reason the report talks about structural friction in specific places rather than producing one number and calling it compatibility.

Values are a structure, not a list of preferences

Schwartz (1992, 2012). The theory of basic human values.

What they did: Surveys of value priorities across dozens of countries and hundreds of samples, testing whether the ten basic values people report fall into a consistent arrangement rather than varying arbitrarily by culture.
What they found: They fall into a circular continuum in which adjacent values are compatible and opposing values are in genuine motivational conflict. Two bipolar axes organise it: Self-Transcendence, meaning universalism and benevolence, against Self-Enhancement, meaning achievement and power; and Openness to Change, meaning self-direction and stimulation, against Conservation, meaning security, conformity and tradition. Opposing positions are not merely different, they pursue incompatible things.
What it changes here: A disagreement across one of those axes cannot be split down the middle, because the midpoint satisfies neither motivation. That is why Structural Fidelity and Values carries the most items of any dimension at fifteen, and why a difference there is reported as macro-scale structural misalignment rather than as something goodwill is expected to absorb later.

Why the body comes first

Of the four dimensions, the physiological one is the only one most people arrive without any vocabulary for, so it is worth spelling out.

Flooding is not a metaphor. Past roughly 100 beats per minute of sustained heart rate, access to language, perspective-taking and memory for what the other person just said measurably narrows, and overlearned responses take over from considered ones. Advice to communicate better fails at exactly that moment, because the capacity it assumes is the capacity that has already gone. This is a physiological state with a recovery time, not a character flaw or a failure of effort.

It is also what makes stonewalling so easy to misread. The partner who goes quiet looks indifferent and is usually the more overwhelmed of the two; the partner who keeps pressing looks aggressive and is usually trying to repair something before it hardens. Each reads the other's way of coping as the attack, which is why the argument escalates without either person intending it to. Naming the mechanism is often the single most useful thing a couple takes from the report.

Two people can hold identical values, want the same future, and still be unable to finish a conversation, if one floods at a volume the other considers an ordinary disagreement. That mismatch is invisible to a quiz about preferences and obvious to one that asks what the body does. It is also the most workable of the four, because the remedy is procedural — pause, recover, return — rather than a matter of changing what either person wants.

The four dimensions

Each partner answers 37 statements alone. The four dimensions are scored separately for each person, and then compared.

Somatic Regulation

6 items

After Gottman and Levenson on conflict physiology

What happens physically when a conversation gets hot. Flooding, shutdown, stonewalling, the urge to pursue or to escape. This is the dimension most people arrive without vocabulary for, and the one that most often explains why two reasonable people cannot finish a disagreement.

What it sounds like

“I go blank and I cannot find the words I had a second ago.”
“I need to leave the room or I will say something I cannot take back.”
“If we stop now it will never get resolved, so I keep going.”

Relational Communion

8 items

The communion axis of the interpersonal circumplex

How openly warmth, affection and reassurance move between two people. Bids for connection and what happens to them: turned toward, turned away from, or missed. The temperature of ordinary days rather than hard ones.

What it sounds like

“I say what I am feeling and it lands somewhere.”
“I show it by doing things, not by saying things.”
“I do not want to need reassurance as much as I do.”

Relational Agency

8 items

The agency axis of the interpersonal circumplex

Who sets direction, who holds limits, who makes the final call. Dominance and deference in the practical sense rather than the moral one: who decides, and how comfortably each person occupies that role.

What it sounds like

“I would rather decide than spend an hour agreeing on where to eat.”
“I will go along with it, and then resent it later.”
“If I do not hold the line, nobody does.”

Structural Fidelity and Values

15 items

After Schwartz on the structure of basic values

What two people actually want a shared life to contain. Money, children, monogamy, ambition, proximity to family, how time is spent. The dimension where agreement cannot be negotiated into existence later, which is why it carries the most items.

What it sounds like

“We have never really said out loud what we each expect.”
“I assumed we wanted the same thing and I am no longer sure.”
“This is the one I do not think I could compromise on.”

Which dimension your situation belongs to

People arrive describing symptoms, not dimensions. These are the four most common descriptions and the dimension each one belongs to, so you can read the relevant part rather than all of it.

Arguments that escalate

Somatic Regulation

Chronic arguments, stonewalling and a hair-trigger temper are usually a regulation problem rather than a disagreement problem.

If this sounds familiar

“We have the same argument over and over and never finish it.”
“One of us shuts down completely and goes silent.”
“It goes from nothing to shouting in about a minute.”
“My heart races and I cannot find the words I had a second ago.”

What that points at

Physiological flooding: diffuse physiological arousal, conventionally marked by sustained heart rate above roughly 100 beats per minute, in which access to language and perspective-taking narrows. Stonewalling is its visible signature and is usually overwhelm rather than indifference. Advice to communicate better fails here, because the capacity it assumes has already gone.

Distance and imbalance

Relational Communion

Coldness, feeling like roommates and carrying all the emotional labour are rarely about whether two people care. They are about whether the caring is arriving.

If this sounds familiar

“We are more like flatmates than a couple.”
“I am the only one who ever plans anything or starts a conversation.”
“I feel invisible, and I do not think it is deliberate.”
“I carry all of the emotional work and nobody notices.”

What that points at

Breakdowns in positive dyadic coping and partner responsiveness: what happens to a bid for connection — turned toward, turned away from, or simply missed — and the baseline temperature of ordinary days rather than hard ones. Whether two people care and whether the care lands are different problems with different remedies.

Money, routine and the future

Structural Fidelity and Values

Friction over money, children or how a weekend is spent is usually a values clash, and it is the one kind of difference goodwill does not absorb over time.

If this sounds familiar

“We cannot agree on money and it comes up constantly.”
“One of us wants children and I am not sure the other does.”
“We want completely different things from a weekend.”
“I assumed we wanted the same life and I am no longer sure.”

What that points at

Structural clashes on the Schwartzian axes: Openness to Change against Conservation, and Self-Transcendence against Self-Enhancement. Opposing positions are in genuine motivational conflict rather than merely different, which is why the midpoint satisfies neither and why splitting the difference does not settle it.

Who decides

Relational Agency

Feeling steamrolled or having no say points at an agency mismatch — and this is the one axis where a difference between two people is often an asset rather than a cost.

If this sounds familiar

“Decisions get made and I find out afterwards.”
“I go along with it and then resent it for a week.”
“I have to decide everything, and I resent that too.”
“I do not get a say in how our life actually runs.”

What that points at

An agency mismatch on the interpersonal circumplex, where agency complements by opposition: two people reaching for the same position collide, and two people both declining it leave decisions unmade. A difference here is not automatically a problem, which is exactly why it is worth measuring rather than assuming.

One situation this is the wrong tool for.

If you are afraid of your partner, or being controlled, monitored, threatened or kept from friends, family or money, that is not a compatibility question and this assessment will not help with it. Some of those experiences sound like the fourth cluster above and are not the same thing. In the UK, the National Domestic Abuse Helpline is 0808 2000 247, free and open 24 hours.

These clusters route a description to the part of the instrument that addresses it. None of them is a clinical category, and nothing here is a diagnosis.

Why difference is not the same as incompatibility

A single similarity score is the wrong shape for this problem, because the same amount of difference means opposite things on different dimensions.

Interpersonal theory, following Carson, holds that some differences between two people invite each other and others collide. A difference in who takes the lead can be the thing that makes a pair work, because agency complements by opposition. A difference in what the two of them want from the next ten years is not the same kind of difference at all, because values conflict by opposition, and no amount of goodwill converts one into the other.

So the four dimensions are not added together. Each is compared in the way that dimension actually behaves between two people, and the result is a position rather than a verdict: where you fit without effort, where the friction is structural, and which of those two a given argument belongs to. A couple arguing about the dishes every week is usually having a regulation problem; a couple arguing about children twice a year is usually having a values problem, and the two need completely different responses.

The published theory above is described in full. Our instantiation of it is not: the specific comparison applied to each dimension, the weighting of items and the thresholds are unpublished, and figures describing them should not be inferred.

Comparative Industry Methodologies

Most compatibility tools measure individuals and infer the relationship. This one measures the relationship directly, because the largest study of the question found that is where the predictive information is: relationship-specific measures accounted for up to 45% of the variance in relationship quality, a person's own traits for 21%, a partner's traits for about 5%, and individual traits added nothing once relationship-level answers were in.

Compatibility methodologies compared with blind dyadic measurement, by unit of analysis and published evidence
Methodology	Unit of analysis	Genuinely useful for	Evidence on couple outcomes	Blind dyadic measurement instead
Individual trait testsMyers-Briggs (MBTI), the Enneagram, the Big Five	One person, measured alone. Compatibility is inferred afterwards by pairing two individual results.	Self-description and a shared vocabulary. The Big Five in particular is a rigorously validated account of individual personality.	Joel, Eastwick et al. (2020), 43 datasets, 11,196 couples: a person's own individual traits accounted for 21% of the variance in relationship quality, and a partner's individual traits for roughly 5%. Once relationship-specific answers were available, individual traits added no further predictive power. MBTI and Enneagram types are also cut from continuous traits at the midpoint, where most people sit, and no published evidence supports type-pair compatibility claims.	Measures both partners separately and blind, then treats the pair as the unit. Relationship-specific measures accounted for up to 45% of the variance in the same study.
Concept frameworksThe Five Love Languages	One person's preferred category for giving and receiving affection.	Starting a conversation about how each person registers affection, often the first such conversation a couple has.	Impett, Park and Muise (2024) reviewed the research against the framework's three core claims, that each person has one primary language, that there are five, and that matching them improves a relationship, and found support for none. Couples whose primary languages matched did not report higher satisfaction. Forced-choice items manufacture a single primary language even for someone who values several equally.	Measures whether warmth actually moves between two people and what happens to a bid for connection. No category match is computed, because matching is the part that did not replicate.
Esoteric and static systemsAstrology, sun-sign compatibility	Each person's date and place of birth: fixed inputs, independent of behaviour.	A low-stakes, playful way to open a conversation about difference.	No peer-reviewed evidence supports sign-pair compatibility. Because the inputs are fixed at birth, the result cannot respond to anything a couple does, learns or changes.	Measures present behaviour under conflict, warmth, decision-making and values. All four can be observed, all four move, and the report says which one to work on.

Scroll sideways to see every column.

The matrix compares unit of analysis, not quality of instrument. Several of these are well built for what they measure. The question it answers is narrower: whether what they measure is where the information about a couple actually sits. On that, the 2020 analysis is direct. Measures of the relationship itself carried the predictive weight, and once those were known, measures of the individuals added nothing.

How this differs from the popular alternatives

Each of these frameworks is doing something real. The question is not whether they are useful but whether their evidence supports the specific job of predicting how two particular people will fare together.

Personality typing: Myers-Briggs, the Enneagram

What it gets right: A memorable description of one person, a shared vocabulary, and the genuine experience of being seen. People remember their type for decades, which no questionnaire manages by accident, and the conversation it starts is often the first structured one a couple has had about difference.
What the evidence shows: Two problems, and the second is the serious one. The first is the bimodal fallacy: the types are presented as two distinct kinds of person, but the underlying traits are continuous and roughly normally distributed, so the cut falls where the most people are and a handful of answers either way reassigns someone. The second is the unit of analysis. Joel and Eastwick measured exactly this question, and individual-difference measures accounted for a fraction of what relationship-level measures did, adding nothing once relationship-level answers were available. No published evidence base supports type-pair compatibility claims for either framework.
What this does instead: Each person is measured separately, and then the pair is the unit. A type describes a person; compatibility is a property of two people, and two accurate descriptions of individuals do not combine into a description of their relationship.

The five love languages

What it gets right: The genuinely useful observation that people register affection differently, and a vocabulary for asking for what you need instead of resenting its absence. For many couples that is the first conversation of its kind they have had, and the framework deserves credit for starting it.
What the evidence shows: Impett, Park and Muise (2024) reviewed the research against Chapman's three core claims — that each person has one primary language, that there are five of them, and that matching them improves a relationship — and found support for none. The matching claim fails most clearly: couples whose primary languages agreed did not report higher satisfaction than couples whose differed. The instrument has a structural problem too: forced-choice items make a person rank expressions of affection against each other, which manufactures a single primary language even for someone who values several roughly equally, and then that artefact is treated as a finding.
What this does instead: Relational Communion measures whether warmth actually moves between two people and what happens to a bid for connection, not whether two people selected the same category from a list of five. No language-match score is computed, because matching is the part that did not replicate.

Dating-app matching algorithms

What it gets right: Genuine value at a problem this does not solve: finding candidates at all. Filtering millions of people down to a plausible few by location, age, stated intent and preference is real work, and no compatibility instrument replaces it.
What the evidence shows: They optimise for the match being made, not for what happens after it. Matching operates almost entirely on stated attributes and preferences — both individual-level data, and Joel and Eastwick found individual traits contributed little once relationship-level information was available. More to the point, the information that predicts best is emergent: how two specific people handle conflict together, whose bid gets turned toward, where their values collide. None of that exists before the two people interact, so no profile-based system can see it, however much data it holds about each person separately.
What this does instead: This starts where matching stops. It assumes two people have already met and asks what happens between them under pressure, which is the information a profile cannot contain and the reason the assessment needs both partners to answer.

Attachment style labels

The internet sorts people into fixed Anxious and Avoidant types. The research treats both as dimensions a person moves along, and what gets called avoidance in a given argument is usually describable as something more specific: a body shutting down, or a bid for connection that stopped landing.

What it gets right: The single most recognisable pattern in relationships: one partner pursues, one withdraws, and each reads the other's response as the problem. Naming that loop helps people stop taking it personally, which is often the first step out of it.
What the evidence shows: The underlying research is dimensional, not categorical. Attachment anxiety and avoidance are continua, measured as such by the standard instruments, and the familiar four boxes are cut from those continua the same way personality types are cut from theirs — the same bimodal fallacy. The dimensions also move: they shift over time and differ between a person's relationships, which a fixed identity cannot express.
What this does instead: Somatic Regulation asks what happens in the body during a specific argument — flooding, shutdown, the urge to pursue or to leave the room. Relational Communion asks what happened to the bid for connection that preceded it. Between them they usually account for what a label was standing in for: withdrawal under physiological flooding, conventionally marked by sustained heart rate above roughly 100 beats per minute, or a breakdown in positive dyadic coping where bids stopped being turned toward. Both are measurable, both move, and both can be worked on. I am an avoidant describes a person and closes the question; I flood at a volume my partner considers normal describes a situation and opens it.

The communication is key myth

Communication is necessary and it is not sufficient. No amount of skill at talking makes two diametrically opposed life values compatible, because the problem is not that the two people have failed to understand each other. Often they understand each other exactly.

What it gets right: A real insight, and the most useful advice available for a large share of relationship trouble. Where a conflict is driven by escalation, by assumptions neither person has checked, or by bids for connection being missed rather than refused, talking differently genuinely resolves it — and that covers a great many arguments.
What the evidence shows: It stops working at the structural boundary. Schwartz showed that values sit on a circular continuum where opposing positions are in genuine motivational conflict: Conservation against Openness to Change, Self-Transcendence against Self-Enhancement. Two people on opposite poles of one of those axes are not miscommunicating. A conversational technique cannot move the midpoint to somewhere that satisfies both, because no such point exists, and framing a structural clash as a communication failure quietly tells both people they are doing it wrong when they are not.
What this does instead: The four dimensions separate the two cases, which is the whole point of measuring rather than advising. A friction on Somatic Regulation or Relational Communion usually does respond to doing things differently. A friction on Structural Fidelity and Values is a question about what each person wants from a shared life, and it needs deciding rather than discussing better. Knowing which kind you are in is more useful than any technique for either.

How the instrument gets checked

The research above is other people's. This part is ours, and it is the reason no accuracy figure appears on this page.

At the end of a paid report, respondents are asked four things: how long they have been together, and three ratings from one to ten — how happy the relationship is, how closely their values align, and how closely they match on handling conflict. Nothing is offered in exchange, which keeps the sample from being bought even though it leaves it self-selected.

The prediction is frozen before any answer can arrive. When a report is generated, the engine version and the dimension scores that produced it are written to an immutable, append-only record alongside it. A rating that arrives months later is scored against what the model actually said at the time, not against whatever the model says now. This is the whole guard against historical overfitting: without a frozen baseline, every improvement to the engine would retroactively appear to have been right all along, because the thing being graded would move each time the grader did.

Append-only matters for the same reason. A respondent who revises an answer adds a row and the earlier one remains, so the record cannot be quietly tidied into a better-looking shape, by us or by anyone who comes after us. The engine version stored with each prediction also means a model change splits the record rather than contaminating it: predictions made by one version are never pooled with predictions made by another.

What that buys is genuine out-of-sample testing. Each frozen prediction is evaluated against an outcome it could not have seen, reported by someone who had no stake in confirming it, which is the only arrangement in which an accuracy figure would mean anything. Refinements to the engine are then tested against predictions held out from the ones used to make the change, rather than against the data that suggested it.

This is necessary rather than decorative. The PNAS analysis found change in relationship quality largely unpredictable from self-report, which means any instrument claiming to forecast a trajectory is claiming more than the best available evidence supports. The only honest way to earn a claim like that is to collect real outcomes against frozen predictions and check.

No validation figure is published yet, and that is deliberate. The sample is small and self-selected: people who bought a report and then chose to answer. A correlation drawn from it would carry more authority than its source deserves, which is the exact error this page criticises elsewhere. When there is enough data to say something defensible, it will be said here with its limitations attached.

What this is not

It assesses behavioural risk factors and structural frictions. It does not issue verdicts, and the distinction is deliberate rather than legal caution.

Not a diagnosis. It is not a clinical instrument and not a substitute for couples therapy. It is a structured way to describe a pattern.
Not a prediction of divorce, and not a timeline. The research it draws on studied couples over years, with measures taken in a laboratory. A questionnaire is not that study. This tool reports no probability that a relationship will end, and no date on which anything happens.
Not a percentage. The compatibility figure is an index out of 100 describing fit across four dimensions. It is not a likelihood of anything.
Not yet validated. The research it is built on is published and peer-reviewed. This particular instrument has not been validated against outcomes, and no accuracy figure for it exists. Collecting what is needed to change that is described above.
Not shared between you. Each partner's individual answers are never shown to the other. Only the dimension scores and the comparison built from them.

The reason for framing the output as risk factors rather than conclusions is practical as well as honest. A reader told what their relationship is tends to reject the label and stop reading, and a verdict leaves nothing to do with it either way. A reader shown a specific friction, named and located, has something to act on. The assessment is only worth anything if it changes what two people do next, which rules out the deterministic phrasing that would make it sound more impressive.

The same reasoning rules out dates. This instrument evaluates statistical trajectories of structural friction — whether the measured gaps between two people point toward friction that recurs or toward a pattern that holds — and it does not predict deterministic timelines, breaking points or break-up dates. The distinction is not a hedge. Joel, Eastwick and colleagues (2020) found that change in relationship quality over the course of a study was largely unpredictable from self-report, across 43 longitudinal datasets. A questionnaire naming the month a relationship will be tested would be claiming something the largest study of its kind could not do.

What the measurement does support is direction. A wide gap on values, or two people who both flood under pressure, are structural facts about a pair that can be read off their answers today, and research associates both with difficulty that recurs rather than resolves. So the report names which of three bands a couple falls in and which specific gap put them there, in language a reader can check against their own experience. It stops short of saying when, because when is the part the evidence does not reach.

Common questions

Short answers to what people ask most, written to stay accurate when quoted on their own.

Is this a personality test?: No. A personality test describes one person. This measures four dimensions for each partner separately and then compares the two sets of answers, because the thing being assessed is the relationship rather than either individual. Joel, Eastwick and colleagues (2020) tested that distinction directly across 43 longitudinal datasets: relationship-specific measures accounted for up to 45% of the variance in relationship quality, a person's own individual traits for 21%, and a partner's individual traits for roughly 5%. Once relationship-specific answers were available, individual traits added no further predictive power.
Does it predict whether a relationship will last?: No, and it is built not to. It identifies behavioural risk factors and structural frictions — patterns that research associates with difficulty — and reports where two people fit easily and where they will have to work. It produces no probability, no forecast and no verdict. The same body of research is the reason: the 2020 PNAS analysis found that change in relationship quality over the course of a study was largely unpredictable from self-report, so any instrument claiming to forecast a trajectory is claiming more than the evidence supports.
What is the compatibility score out of 100?: It is an index describing fit across four dimensions, not a percentage likelihood of anything. A score of 72 does not mean a 72% chance of staying together, and it is not a grade. It is a position: a summary of where two people's answers align without effort and where they diverge in ways that tend to cost something. The detailed report is more useful than the number, because it names which specific frictions produced it.
How is this different from Myers-Briggs or the Enneagram?: Those assign one person a type, and types are cut from continuous distributions at the midpoint — where most people actually sit — so small answer changes reassign someone. The deeper difference is the unit of analysis. A type describes an individual, and compatibility is a property of a pair; two accurate descriptions of two people do not combine into a description of their relationship. No published evidence base supports type-pair compatibility claims for either framework.
What about the five love languages?: Impett, Park and Muise (2024) reviewed the evidence against Chapman's three core claims — that each person has one primary language, that there are five, and that matching them improves a relationship — and found support for none of them. The matching claim fails most clearly: couples whose primary languages agreed did not report higher satisfaction than couples whose differed. What does hold up is simpler — receiving expressions of affection predicts satisfaction, in whatever form it arrives. This assessment computes no language-match score.
Is my partner avoidant?: Probably a less useful question than it feels like. The internet sorts people into fixed Anxious and Avoidant types; the research measures attachment anxiety and avoidance as dimensions a person moves along, which shift over time and differ between that person's relationships. What gets called avoidance during a specific argument is usually describable as something more concrete: withdrawal under physiological flooding, where sustained arousal narrows access to language, or a breakdown in positive dyadic coping where bids for connection stopped being turned toward. Both are measurable and both change, which a fixed label cannot express. Somatic Regulation and Relational Communion measure them separately.
Isn't communication the key to a good relationship?: Communication is necessary and it is not sufficient. Where a conflict is driven by escalation, unchecked assumptions, or bids for connection being missed rather than refused, talking differently genuinely resolves it, and that covers a great many arguments. It stops working at the structural boundary. Schwartz showed that values sit on a continuum where opposing positions are in genuine motivational conflict, such as Conservation against Openness to Change. Two people on opposite poles are not misunderstanding each other; often they understand each other exactly. No conversational technique moves the midpoint to a place that satisfies both, because no such place exists. The useful step is knowing which of the two kinds of friction you are in.
Can it tell me when my relationship will fail?: No, and it is built not to. The assessment evaluates statistical trajectories of structural friction — whether the measured gaps between two partners point toward friction that recurs or toward a pattern that holds — and reports one of three directional bands. It names no date, no breaking point and no probability. Joel, Eastwick and colleagues (2020) found change in relationship quality largely unpredictable from self-report across 43 longitudinal datasets, so a questionnaire naming a month would be claiming more than the largest study of its kind could deliver. What the report does say is which specific gap puts a couple where they are, which is checkable.
Has this particular instrument been validated?: Not yet, and no accuracy figure for it is published. The research it is built on is peer-reviewed; this specific implementation has not been tested against outcomes. Validation work is underway: predictions are frozen in an immutable snapshot when a report is generated, and respondents are later asked about real outcomes, so each prediction can be scored against what the model said at the time rather than a later version of it. No figure will be published until the sample supports one.
What happens to the answers?: Each partner's individual answers are never shown to the other. Only the four dimension scores and the comparison built from them are shared between two people in a couple's report. The privacy notice describes what is stored, for how long, and what is shared with any third party.
How long does it take?: About four minutes. Each partner answers 37 statements alone, on their own device, without seeing the other's responses. The comparison needs both sets, so the report appears once the second person has finished.

Sources

Joel, S., Eastwick, P. W., and 83 co-authors (2020). Machine learning uncovers the most robust self-report predictors of relationship quality across 43 longitudinal couples studies. Proceedings of the National Academy of Sciences, 117(32), pages 19061 to 19071. doi 10.1073/pnas.1917036117
Levenson, R. W., & Gottman, J. M. (1983). Marital interaction: Physiological linkage and affective exchange. Journal of Personality and Social Psychology, 45(3), pages 587 to 597.
Levenson, R. W., & Gottman, J. M. (1985). Physiological and affective predictors of change in relationship satisfaction. Journal of Personality and Social Psychology, 49(1), pages 85 to 94.
Gottman, J. M., & Levenson, R. W. (2000). The timing of divorce: Predicting when a couple will divorce over a 14 year period. Journal of Marriage and the Family, 62(3), pages 737 to 745.
Buehlman, K. T., Gottman, J. M., & Katz, L. F. (1992). How a couple views their past predicts their future: Predicting divorce from an oral history interview. Journal of Family Psychology, 5(3/4), pages 295 to 318.
Gottman, J. M. (1994). What Predicts Divorce? The Relationship Between Marital Processes and Marital Outcomes. Lawrence Erlbaum Associates.
Bakan, D. (1966). The Duality of Human Existence: An Essay on Psychology and Religion. Rand McNally. The origin of agency and communion as the two fundamental modalities.
Carson, R. C. (1969). Interaction Concepts of Personality. Aldine. The interpersonal circumplex and the principle of complementarity.
Schwartz, S. H. (1992). Universals in the content and structure of values: Theoretical advances and empirical tests in 20 countries. Advances in Experimental Social Psychology, 25, pages 1 to 65.
Schwartz, S. H. (2012). An overview of the Schwartz theory of basic values. Online Readings in Psychology and Culture, 2(1).
Brennan, K. A., Clark, C. L., & Shaver, P. R. (1998). Self-report measurement of adult romantic attachment: An integrative overview. In J. A. Simpson & W. S. Rholes (Eds.), Attachment Theory and Close Relationships, pages 46 to 76. Guilford Press.
Impett, E. A., Park, H. G., & Muise, A. (2024). Popular psychology through a scientific lens: Evaluating love languages from a relationship science perspective. Current Directions in Psychological Science. doi 10.1177/09637214231217663

Four minutes, answered separately

You and your partner each answer alone, on your own devices. The comparison is built from both sets of answers.

Start the free test