I keep staying, even though I want more.

Attachment research began with John Bowlby and Mary Ainsworth, who observed that children develop expectations about whether closeness is available and whether distress will be responded to. Later researchers, including Cindy Hazan and Phillip Shaver, extended this to adult romantic relationships.

What this line of research reasonably supports is that people carry working expectations about closeness, rejection, reassurance and availability. What it does not support is treating an attachment style as a fixed personality type or a diagnosis. Longitudinal studies find meaningful continuity in attachment patterns, and also meaningful change, particularly following significant relationships, life transitions and therapy.

So attachment is best used here as a description of a pattern you may recognise, not a label you are stuck inside.

Alongside attachment, ordinary learning matters. Relationships teach people what tends to follow what: that asking directly leads to withdrawal, that going quiet restores warmth, that being useful reliably produces approval.

Behavioural science calls this instrumental or operant learning, described by Edward Thorndike and later B. F. Skinner. Behaviour that appears to produce a desired outcome, or to reduce a feared one, tends to be repeated. Albert Bandura's work added that people also learn by watching how those around them handled closeness, conflict and repair.

This is a strong and unglamorous explanation. Not that you are drawn to difficulty, but that you have practised particular responses many times, and practised behaviour becomes fast, automatic and easy to reach for.

One finding in behavioural science is unusually robust: behaviour maintained by unpredictable reward is more persistent than behaviour maintained by consistent reward. Ferster and Skinner mapped these schedules in detail, and variable schedules produce steady, hard-to-extinguish behaviour.

Applied carefully, this helps explain something people find bewildering about themselves. If affection arrives unpredictably, checking, waiting and hoping are all repeatedly, occasionally rewarded. The behaviour persists precisely because you cannot predict when the next warm moment will come.

A caution, because this idea is widely misused. These are findings about schedules of reinforcement, largely from controlled settings. They do not establish that anyone is chemically addicted to a partner, and they do not make inconsistency in a relationship a clinical condition. They describe why a pattern is hard to stop, not what kind of person you are.

Neuroscience can say something modest and genuinely useful here. Reward-learning research, including work by Wolfram Schultz, shows that brain activity tracks prediction error: the gap between what was expected and what occurred. Anticipation, rather than the reward itself, drives a great deal of learning, and uncertainty amplifies that anticipatory response.

That offers a careful account of why the almost, the maybe and the nearly can feel more absorbing than something settled. Uncertainty keeps a prediction open, and an open prediction holds attention.

What this research does not license is a story about one neurotransmitter making you love the wrong person. Human relationships involve memory, meaning, context, culture and choice, and mapping them onto a single chemical is not something the evidence supports.

Aaron Beck described how people hold core beliefs that shape what they notice and how they interpret it. Jeffrey Young extended this into schema therapy, identifying long-standing patterns with names like emotional deprivation, subjugation, self-sacrifice and abandonment.

In relationships these beliefs often sound like: love has to be earned. My needs are too much. People leave. I have to be useful to be loved. If I stop trying, I'll lose them.

Beliefs like these rarely announce themselves. They operate as a filter, so that a partner's brief attention reads as proof of love while months of absence read as circumstance. Naming the belief is what makes the filter visible, and a visible filter can be questioned.

Familiarity does real work. Robert Zajonc's research on mere exposure found that repeated exposure tends to increase liking, and social-cognition research finds that people process familiar patterns more fluently, with less effort and less alarm.

That is a more defensible account of repetition than the popular claim that people unconsciously seek out their childhood wounds. Some clinicians do describe repetition compulsion, a psychoanalytic idea rather than an established empirical finding, and it should be treated as one interpretation among several.

What can be said is this: a dynamic you already know how to survive asks less of you than one you have never practised. Recognising something is not the same as it being good for you.

The adult brain remains capable of learning and adaptation throughout life. That is well established, and it is the honest basis for hope here.

It is also routinely oversold. Learned patterns generally are not deleted. Extinction research by Mark Bouton and others suggests that old learning tends to remain available, particularly under stress or in familiar contexts, while new learning is added alongside it. This is why a pattern you understand well can still reappear on a bad night.

Which is why four weeks is described here as a beginning. The realistic goal is to notice an old response early enough to make a different choice, and then to repeat that different choice often enough that it stops feeling foreign. Where distress is severe or long-standing, evidence-based therapy with a qualified professional is the appropriate route, and this programme is not a substitute for it.