How LoveScore calculates compatibility

LoveScore is a couple compatibility test (lovescore.love). It is not a name-based “love calculator” and not fortune-telling: behind the result is a deterministic algorithm. The same answers always produce the same result — down to the exact point. AI only writes the narrative of your report and never affects the numbers.

An algorithm, not fortune-telling

Services that “calculate love” from names or birth dates conjure a number out of nothing: their input has no connection to your relationship. LoveScore works differently. Each of the two partners answers 47 questions about how your particular couple works — separately, from their own device — and every score is computed by fixed rules, the same for every couple and described step by step below. There is no randomness in the calculation, no “magic” and no neural network: a language model joins only at the very last step, to phrase the text of the report from numbers that are already final.

Your answers are never revealed to your partner: each of you sees the couple's shared report, but not the other person's line-by-line answers. The test takes about 7 minutes.

How the questionnaire is built

The 47 screens are 45 substantive items (9 dimensions, 5 items each) plus 2 built-in attention checks. Each item is a short statement you rate on a scale from 1 to 5, from “strongly disagree” to “strongly agree”.

Some items are direct (agreeing raises the dimension's score), others are reversed (agreeing lowers it). Reversed items are not there to confuse you. First, they catch autopilot: if you click the same answer without reading, direct and reversed items start to contradict each other — and the system notices. Second, they protect against acquiescence, the habit of agreeing with any nicely worded statement. To score high, it is not enough to nod along: you have to answer what is actually being asked.

The 9 relationship dimensions

Each dimension has its own page with a detailed description; all nine are collected in the dimensions glossary:

• Trust — confidence in your partner's honesty and good will.
• Communication — whether hard conversations actually reach a resolution.
• Empathy — how accurately you read each other's feelings.
• Emotional Maturity — the cost of conflict: a pause instead of escalation.
• Responsibility — promises, everyday life and your share of joint decisions.
• Peace of Mind — a reversed scale: peace of mind instead of jealousy and control.
• Togetherness & Space — a match scale: aligned needs matter, not their level.
• Future Vision — a match scale: whether your pictures of a shared future coincide.
• Partner Responsiveness — how well your partner hears you; not part of the Love Score.

Partner Responsiveness is the only dimension where you rate not yourself but how well your partner hears you. That is why it is not part of the overall Love Score: otherwise one person's contribution would be counted twice. Instead, it powers the comparison of “how I see myself” with “how my partner sees me”.

How the Love Score is calculated

Step 1. From answers to a 0–100 score

Each answer is first brought to a single direction: for direct items we take the answer itself, for reversed ones — its mirror, because agreeing with a reversed statement means exactly what disagreeing with a direct one means. The item scores within a dimension are then averaged and carried linearly onto a 0–100 scale.

The transfer is linear, with no tweaks inside the scale: all five items count equally, none is “more important” than the rest, and the same shift in answers moves the score by the same amount wherever on the scale it happens. Extreme answers across all five items give 0 and 100; the neutral middle answer lands exactly on 50. Zero and a hundred are the edges of the questionnaire itself, not a person's place among other people: no population takes part in the calculation anywhere.

Step 2. Two personal scores — one couple score

For most dimensions, the couple score follows the “level” rule, and it is assembled from two quantities at once: the couple's average and the score of whichever partner sits lower.

Why not a plain average? An average would equate a couple where the dimension sits mid-range for both with a couple where one is at the top and the other at the bottom — and those are fundamentally different situations: in the second one, one of the two lives without that support entirely. So the couple score counts both the average level and the minimum: the average weighs more, but the weaker link weighs enough that the stronger partner's resources cannot fully “cover” the other's deficit. A dip in one partner cannot be bought out by the other — only softened. With equal partners, the rule reduces to their shared level.

Match scales: alignment matters, not level

For Togetherness & Space and Future Vision, “higher = better” simply does not exist: a couple where both value separateness and a couple where both want closeness are equally aligned. So exactly one quantity enters the calculation — the distance between the two partners' scores: the closer the answers, the higher the couple score. The average level plays no part at all.

The score falls as the gap between partners grows, and at a divergence that no longer means different preferences but mutually exclusive life plans it drops to zero. A plain average would be dangerous here: a couple with one partner at one pole and the other at the opposite one would get a reassuring middle that masks a structural conflict.

Two scales — Communication and Responsibility — are hybrid: the main contribution comes from the level, and a smaller share of the score reflects how closely the partners' pictures match. A large gap in answers about communication is a signal in itself: two people are describing the same process differently. And a persistent imbalance in who carries the household cannot be reduced to an average level of responsibility.

Why Peace of Mind is a reversed scale

The items of this dimension ask about jealousy and the need for control — phrased affirmatively, with no double negatives. The raw score therefore means “higher = more jealous”, and before the couple-level aggregation it is flipped: the less jealousy and need for control the answers carry, the higher the score. Publicly, the dimension appears as Peace of Mind, where higher means calmer. Importantly, mutually high jealousy does not count as “compatibility”: mutual control is a risk configuration, not agreement — which is why this is a level scale, not a match scale.

The final score

The Love Score is a weighted sum of the couple scores of eight dimensions (Partner Responsiveness excluded). Trust and Communication carry the most weight: trust is the precondition for everything else, and communication is the only dimension that describes the process of resolving disagreements rather than a trait. Match scales weigh less: an aligned couple gets their maximum from them anyway.

On top of the sum sits a non-compensatory cap: if the couple has a deep dip in one of the critical dimensions (trust, communication, peace of mind — and, for couples with history, future vision: at the very start of a relationship a mismatch of plans does not trigger the cap), the final score is limited from above. Excellent answers in the other dimensions cannot “buy” a high Love Score around the dip — otherwise a couple with broken trust and beautiful communication would see “you are doing great”, and that would be misinformation.

Answer quality control

The result is only meaningful when the answers are attentive, so several independent checks run alongside the calculation:

• Attention checks — two service questions with a known correct action, built into the questionnaire.
• Inconsistency index — several pairs of items ask about the same thing in opposite phrasings; a strong contradiction within such a pair means the answers cannot be taken at face value.
• Acquiescence detector — compares answers to direct and reversed items: someone who agrees with everything produces a telltale skew.
• Pace — answers that come too fast and a session that is too short overall, as well as long streaks of identical answers.

We deliberately do not publish the exact thresholds — they serve the system, not those who want to game it. Here is how it works: with moderate doubts, the report is still built but flagged — exact scores are replaced with ranges, and the wording becomes more cautious. With clear unreliability (for example, both attention checks failed), no report is built at all: the service offers to retake the test, because an honest report cannot be made from random answers.

Partner archetypes

Beyond the numbers, each partner receives an archetype — a portrait assembled from the profile of their answers. There are 12 archetypes: eight are built on a lead dimension (for example, Open Book — trust, Diplomat — communication), three on characteristic combinations of two dimensions (Warm Hedgehog, Open Sky, Lightning Rod), and one — Steady Flame — goes to an even profile with no pronounced lead dimension. No archetype is better or worse than another: it is a language for talking about differences, not a ranking. Read more on the archetypes page.

What AI does — and what it does not

The separation of roles is strict. The deterministic core computes every score, bar and flag — with no language model involved. The model receives the finished numbers and turns them into a coherent report text. It cannot influence the scores by construction: they are computed before it is ever called.

Every generated text passes an automatic validator. A stop list bans predictions (“you will break up”, “probability of divorce”), clinical labels and diagnoses, and false normativity such as comparisons with “most couples” — no basis for such comparisons exists. A separate check verifies every number in the text against the core's calculation: the model is not allowed to “invent” a score. If the text fails the checks after repeated attempts, a deterministic template built from the same numbers is shown instead — the service never displays unverified text.

The method's limits — honestly

We believe the boundaries of applicability should be named outright:

• It is self-report. The test measures how you describe your relationship, not the relationship itself. The mood of the day, the wish to look a little better and different readings of the wording all affect the answers.
• It is not a clinical instrument. LoveScore has not undergone clinical validation, makes no diagnoses and does not assess mental health.
• It is not a prediction. No score has been validated against real relationship outcomes. The Love Score is an index of how consistent your answers are with the test's model — not a forecast of the future and not a comparison with other couples.
• It is not a substitute for a professional. If your relationship is in crisis or you feel unsafe, a report will not help the way a qualified human will.

Within these boundaries, the test does what it promises: it gives a couple a structured occasion to see their strengths and their points of tension — and to talk about them in words.

LoveScore is an entertainment and educational service for adults 18+. The results are not psychological advice, a diagnosis or a prediction of your relationship's future.