The loop, one turn at a time
Something registers as distance. A shorter reply, a cancelled evening, an unexplained mood.
The anxious response is to close the gap: ask, check, press for a clear answer, ask again in a different way. From inside it is not pressure, it is trying to find out whether things are all right.
That arrives on the avoidant side as demand. The deactivating response follows: answer later, answer briefly, need some room, handle it alone. From inside that is not punishment, it is trying to get the volume down.
Which is, precisely, more distance. Which is the trigger the loop started on.
Each turn costs a little more than the last, because each one adds evidence. The anxious partner now has a record of withdrawal, so smaller signals count. The avoidant partner now has a record of pressure, so smaller requests count. Nothing new needs to happen for the third turn to be worse than the first.
Why it starts at all
Two things, and neither of them is bad judgement.
The first is that each person’s behaviour confirms the model the other already carries. Someone who half expects to be left meets someone who does leave the room. Someone who half expects to be crowded meets someone who does close in. Confirmation arrives fast, and being confirmed is easily read as being understood.
The second is that the loop generates intensity, and intensity is genuinely difficult to distinguish from depth of feeling while you are inside it. Relief when contact resumes is a strong feeling. It is also a feeling produced by the gap that preceded it.
Two positions, not two labels
Here is what the label framing hides.
Take a pair where one sits slightly above the middle on anxiety and the other slightly above the middle on avoidance. Both will be handed their labels by any quiz. The loop between them barely turns: it takes a real event to start, and it stops on an ordinary conversation.
Now take a pair far out on each axis. The same labels, and the loop starts on a delayed reply and needs an argument to stop.
Same two words, a different relationship. This is the reason to know your position on each axis and not just which corner you rounded into, and it is the reason the pillar page spends its length on the axes rather than on the four portraits.
There is a second thing labels hide, worth naming because it changes what the loop looks like. Someone in the fearful corner is high on both axes, so they can run both halves of the cycle themselves, pursuing one week and withdrawing the next. Read as a label they look like the anxious partner. Read as two positions, they are a different situation entirely. The avoidant attachment page sets that distinction out.
Where it breaks
Not, mostly, by matching better. The research on couples is consistently less interested in whether two people are well paired than in what they do after something has gone wrong. Rupture is close to universal. What varies is whether there is a route back.
Concretely, and this is the part that generalises:
- Name the loop, not the person. “This is the thing that happens to us” is workable. “You always shut down” is a turn of the loop.
- Agree what a pause is, in advance. The whole cost of a withdrawal, for the anxious side, is not knowing whether it ends. A pause with a stated end is a different object from a pause without one.
- Treat coming back as the event. If the return gets met with an account of the absence, the return becomes expensive, and expensive returns get shorter over time.
- Move a position rather than win a turn. Every one of these acts on the axis, not on the argument. That is why it is slow, and why it lasts.
Couples therapists work on exactly this material. A page is a description; the work is not something a description does.
What this says about compatibility
Less than the calculators suggest, and something more useful than nothing.
What the evidence broadly supports is that similarity of personality traits predicts fairly little about how satisfied two people end up, while how each person handles closeness, conflict and repair predicts more. That points compatibility away from matching two profiles and toward reading what each pair’s two positions will do to each other, which is a different question with a different answer.
Noetly measures the two attachment axes, each with its own margin of error, alongside four more dimensions of how you approach a relationship and the rest of your profile. Comparing two profiles is the next thing being built rather than something shipped today, and the honest version of it will describe the likely dynamic and its pressure points, not return a percentage.