All articles ←

The two things not measured by comparison: attachment and stability

9 September 2026·8 min read·By The Zugot Project team

You can be beautifully aligned on values, interests, age and lifestyle and still fail, because the way each of you responds when the relationship is under strain is a dimension none of those comparisons captures.

A dimension with no "alike" and "different"

Almost everything a matching system measures is relative. The values score is the gap between two positions. The interests score is the size of the overlap. The age score is the distance in years. In each case, "how many points does this candidate deserve" has no answer without knowing who the other person is.

Two dimensions don't work that way. In one — emotional stability — two stable people are the best combination, however alike or unalike they are. In the other — attachment style — what decides is not the similarity between the sides but whether their combination produces a loop that feeds itself.

Emotional stability — the strongest predictor

Of all the single traits measured in the literature on relationship satisfaction, emotional stability is the one that recurs most consistently as a predictor. Not charisma, not intelligence, not shared interests — a person's capacity to return to equilibrium after an encounter with stress, disappointment or conflict.

The reason makes sense once you consider what a relationship actually demands. A relationship is a long sequence of small events, each requiring regulation: a sick child, an argument at an inconvenient hour, an unintended slight, financial pressure. A stable person absorbs the event and recovers. A less stable person turns each of them into an incident requiring management, and their partner becomes a full-time crisis manager.

So the score on this item is computed differently from every other: not as the distance between the two, but as the average of both their stability levels. Two stable people receive full marks. Two unstable people receive little — not because they aren't alike, but because they are alike in precisely the thing where being alike doesn't help.

Attachment: anxiety and avoidance

Attachment theory describes a person's relational patterns along two independent axes. The first is anxiety: how troubled someone is by the possibility that a partner will withdraw, and how much reassurance they need to feel settled. The second is avoidance: how much someone avoids closeness, struggles to lean on another person, and prefers distance even when the relationship is good.

Secure attachment — low anxiety, low avoidance — is among the most replicated predictors in the literature for satisfaction and felt safety in a relationship. And like emotional stability, it isn't a question of similarity: two secure people are an excellent combination, but two anxious people are not an excellent combination merely because they are both anxious.

The combination the system flags as problematic is the classic anxiety-avoidance dynamic: one side seeking closeness and reassurance, the other withdrawing the more is asked of them. Every request intensifies the withdrawal, and every withdrawal intensifies the request. Both sides are behaving entirely reasonably from their own vantage point, and the loop tightens all the same.

Accordingly, these two combinations are handled separately in the score: a pairing of two secure people receives a boost, and one that produces the anxiety-avoidance loop receives a deduction. Both sit outside the weighted calculation rather than inside it, precisely because they aren't measuring similarity.

Why the questionnaire here is deliberately short

The attachment screen contains just two items — one per axis. That is very little in psychometric terms, and the decision was made with open eyes.

The cost is clear: a single item places a person on an axis coarsely, not precisely. The gain is that people actually finish. A full attachment inventory runs to dozens of items, and it arrives at a point where the registrant has already answered dozens of other questions; what happens in practice is that people abandon it or answer carelessly, and both are worse than a coarse but honest measurement.

One item was also dropped entirely — "I find it difficult to depend on others" — because it measured avoidance from the less useful direction: it asked how a person theorises about dependence rather than what they actually do inside a relationship. Another item, about the need for reassurance, wasn't dropped but promoted: it became a standalone question with its own matching rule, rather than being asked twice on the same form.

What to do with this

If you know you sit at the anxious end of the axis, the most useful information isn't to look for someone who will love you enough that the anxiety disappears. Anxiety doesn't dissolve under sufficient affection. What does matter is a partner whose presence is steady and predictable — someone who doesn't vanish when things get hard, and whose response to distress is to move closer rather than to close up.

And if you sit at the avoidant end, the question worth asking about a candidate isn't how attractive she is but how much space she needs. Someone who requires frequent reassurance won't "learn to give you room" — she will experience the room you take as withdrawal, and that will intensify exactly what you're trying to reduce.

For both ends: these are traits that can be worked on, and the work succeeds. But it is work done inside a relationship, not before it — so the initial match needs to be one that makes that work possible, rather than one that produces the loop that makes it impossible.