What the research says
In 2020, Samantha Joel and around a hundred co-authors pooled 43 longitudinal studies of couples and used machine learning to find which variables best predicted relationship quality. The result, published in PNAS, is clear on two counts.
Variables belonging to the relationship predict well: reported satisfaction, perceived commitment, felt appreciation, conflict. Individual characteristics of each partner predict much less. And similarity between partners adds almost nothing once the rest is accounted for.
Put differently: the question compatibility tests ask, by comparing traits, lands on the least predictive dimension available.
Why percentages make no sense
A compatibility percentage assumes there is a known function mapping two profiles onto a probability of success. No such function exists in the literature. What does exist are partial associations between certain configurations and certain difficulties.
There is a measurement problem on top. Most compatibility tests measure each partner with a dozen questions and no margin of error. Combining two imprecise measurements produces something less precise than either, not more.
The three comparisons that hold
Three gaps between profiles are documented as sources of identifiable tension.
The attachment configuration. This is the best established one. High anxiety and high avoidance both predict lower satisfaction, for the person carrying them and for their partner, and some combinations produce dynamics as well known as the anxious-avoidant loop.
The gap in values. Values bear on concrete, repeated trade-offs: money, time, children, the place work takes. Two people who rank security and stimulation differently are not arguing about an abstraction. They are arguing every six months about a decision.
The gap in disposition to commit and in sociosexuality. These two dimensions describe what each person expects the shape of the relationship to be. A wide gap condemns nothing, but it turns into an explicit negotiation rather than a tacit understanding.
What Noetly will do, and will not
Comparing two profiles does not exist yet: it requires both partners to have one, properly measured. That is the next piece of work on the love profile.
What it will return: both partners’ positions on each axis with their precision, the configurations known to produce named tensions, and the sharpest gaps in values.
What it will not return: a compatibility score. The data does not support one, and an invented number would do more damage than it would answer.
Until then the first step is the same for everyone: having a profile to compare.