The Architecture of Connection (Beyond Attachment Styles, Part III): Towards a Neurodivergent Understanding of Relational Security

In my previous two articles, I explored why standard attachment categories, secure, avoidant, anxious, disorganised, often fail to capture the relational experience of autistic, ADHD, and AuDHD adults. In Part I, I introduced the ideas of relational accessibility, authentic security, and contextual attachment as ways of describing connection that shifts with sensory state, masking demand, and environment, rather than remaining fixed as a trait. In Part II, I applied this lens to the psychometric tools themselves, showing how the same questionnaire item can mean entirely different things depending on the respondent's neurotype.

At the end of that piece, I said I would return to something earlier and, in some ways, more foundational: how these patterns begin. Not the adult presentation, but the developmental process that produces it.

Mirroring, and what happens when it fails

A great deal of early relational development rests on mirroring: a caregiver reflecting a child's internal state back to them, accurately and in a form the child can recognise. Reflected enough times, in a form the child can access, this becomes the basis for a stable sense of being understood, and, eventually, the basis for a stable sense of self.

For many neurodivergent children, this process is disrupted early and often, not through neglect, but through mismatch. A caregiver may soothe with words when the child needs stillness. A caregiver may read distress correctly but respond with a regulation strategy built for a neurotypical nervous system, more eye contact, more verbal reassurance, more physical closeness, each of which can increase rather than reduce distress for a sensory-sensitive child. The caregiver's read of the emotion may even be right. The response built on top of it is calibrated for a different kind of nervous system.

This is mirroring failure without malicious intent on either side. It is also, over enough repetitions, developmentally significant. A child whose internal states are consistently misread does not simply miss out on comfort in the moment. They begin to internalise something about the reliability of being understood at all.

From misattunement to expectation

This is the mechanism I want to name clearly: repeated misattunement does not stay in the past. It becomes encoded as an expectation, carried forward into adult relationships, often below conscious awareness.

An adult client who braces before asking for something they need is not usually reacting to the present relationship. They are often reacting to a well-founded, historically accurate expectation that needs, once expressed, will be met with a mismatched response. The bracing is not disproportionate. It is a reasonable output of a well-calibrated internal model, calibrated against real data, just data that is decades old and may no longer describe the person in front of them.

This is also, I think, where masking begins in its earliest form. Long before it becomes the conscious, effortful camouflaging I described in Part I, masking often starts as a child's early adaptation to being misread: presenting the version of distress, or need, or interest, that seemed most likely to be understood, rather than the version that was actually true. What later looks like a personality trait, guardedness, self-sufficiency, difficulty asking for help, is frequently the adult residue of a very old, very sensible strategy.

Why "relational security" needs a different shape

Bowlby and Ainsworth's model of security rests heavily on proximity-seeking: does the child seek the caregiver when distressed, and is that seeking met. This is a reasonable proxy in the population it was built on. But if connection is only measured by whether someone seeks it in a recognisable way, at a recognisable moment, we will consistently misread neurodivergent security as its absence.

I want to propose a different way of asking the question. Rather than does this person seek proximity when distressed, I think the more accurate clinical question is: what structural conditions need to be in place before connection becomes accessible to this person at all?

I am calling this the architecture of connection: not a style, and not a fixed internal state, but a set of load-bearing conditions that either support or undermine a person's capacity to feel safely connected. For many neurodivergent clients, these conditions include predictability of contact, explicit rather than implied communication, a regulated sensory environment, and permission for contact to be intermittent without being read as absence. Remove any one of these, and connection can become inaccessible, not because the bond has weakened, but because the structure holding it up has given way somewhere.

This builds directly on the terms from Part I. Relational accessibility describes the moment-to-moment experience. Contextual attachment describes the fact that this shifts with environment. Architecture, I think, describes the underlying structure that determines which contexts will support accessibility and which will not, the thing we would actually need to build, or rebuild, in therapeutic work.

A note on Neurodivergent Attachment Theory

Montsheng Letsoalo's Neurodivergent Attachment Theory (2025), which I referenced in Part I and Part II, proposes five attachment styles that map onto some of this structure directly: Sensory-Secure, Masking-Avoidant, Hyperfocus-Attached, Looping-Disorganised, and Cognitive-Connector. I find this a useful complementary taxonomy, particularly Sensory-Secure and Masking-Avoidant, which correspond closely to what I am describing here as architectural conditions being present or withdrawn. NAT itself acknowledges that it does not yet fully account for cultural and intersectional variability in attachment, a limitation worth holding in mind when applying any of these frameworks clinically.

Clinical implications

In practice, this reframing changes what we are listening for. Instead of asking whether a client's presentation fits a recognisable attachment category, I am asking what conditions, if restored or introduced, would make connection more accessible to them. Often the answer is strikingly concrete: a specific communication pattern, a way of signalling availability without demanding immediate response, an agreed script for repair after rupture that does not rely on real-time verbal processing.

In couples work, this is frequently where the most movement happens. Once both partners can name the specific structural conditions each of them needs, rather than diagnosing each other's attachment style, ambiguity tends to reduce sharply. The work becomes architectural rather than interpretive: what do we need to build, together, for connection to be reliably accessible to both of us.

Looking ahead

Having traced this from early mirroring through to an adult structural model, my next piece will turn to repair: what rupture and recovery actually look like when one or both partners process rupture on a different timeline, and why the standard expectation of immediate verbal repair can itself become a structural failure point.