A Predictive Processing Account of Relational Trauma
The Relational
Safety Framework
Understanding how early relationships become learned models of what connection is likely to mean and cost, and how present evidence can confirm, challenge, or revise those predictions.
Developed by Stephanie Underwood, MSW, RSW
Explore the frameworkThe Core Idea
Safety is an inference,
not a signal
Most trauma models were built to explain what happens after a single overwhelming event — an accident, a natural disaster, an act of violence. But many people carry the effects of something different: relational trauma.
Relational trauma rarely comes from one moment. It develops through repeated experiences of rejection, unpredictability, emotional unavailability, coercion, or threat within the relationships on which a person depended, often beginning in childhood.
RSF proposes that relational safety is not detected by a separate bodily system, and it is not contained inside a tone of voice, facial expression, or racing heart. The brain must infer what current signals mean by integrating sensory evidence with context, memory, prior learning, and the predicted consequences of available actions.
Early relationships repeatedly calibrate those models. When care is inconsistent, needs are punished, or repair is unavailable, the brain may assign greater probability to rejection, helplessness, or relational cost in similar situations later on. These predictions are adaptations to experience, not proof that a person is broken.
The body remains part of the loop. Autonomic and endocrine responses change heart rate, breathing, muscle tension, energy use, and other internal conditions. Those changes return to the brain as interoceptive evidence and can influence the next inference. The body participates, but it is not an archive where trauma is stored.
No relational cue is universally safe or threatening. Its meaning depends on learning, context, cue reliability, available actions, social resources, and predicted cost.
"Relational distress is maintained not because the body keeps a score, but because the brain keeps predicting from what it learned."
— The Relational Safety Framework
The Predictive Processing Loop
Threat is inferred, not simply detected
There is no single safety detector and no universal pathway from a cue to a response. RSF describes a distributed, recurrent process in which present evidence is interpreted through learned models and the predicted consequences of possible actions.
Current evidence
Social and environmental cues arrive alongside interoceptive signals from the body. These signals are evidence, not meaning by themselves.
Learned model and context
Prior relational learning, memory, context, and confidence in the available evidence shape which explanation seems most plausible.
Threat inference and action
The brain estimates likely harm, uncertainty, cost, and available control, then organizes attention, physiology, and protective action.
Feedback and updating
The response changes the body and the relationship. The observed outcome becomes new evidence that can confirm, refine, or revise the model.
Physiology is part of the loop. It is an organized response and a new source of interoceptive evidence, not the system that independently decided what the situation meant.
How We Assess Safety
Four information classes that shape relational safety appraisal
RSF proposes that these four interacting sources help shape inferences about relational threat and safety. They are not four independent brain systems, and safety is not a single switch or bodily state.
Predictability
What is likely to happen next, and how certain is that estimate? Uncertainty can increase monitoring and anticipated cost. Predictability alone is not safety because harm can also be perfectly predictable.
Controllability
What can I do, and are my actions likely to change the outcome? Boundaries, exit, repair, help-seeking, and influence alter the predicted consequences of staying engaged.
Relational Signaling
What does this person's behaviour mean here? Tone, expression, timing, silence, and responsiveness are learned, context-sensitive evidence. No cue is inherently safe or dangerous.
Interoceptive Cost
What is my body signaling now, and what physiological or emotional demand do I predict if I stay engaged? Current bodily state and anticipated regulatory cost can alter appraisal and action.
Safety becomes more likely when expected harm and regulatory cost are acceptably low given current evidence, prior learning, available actions, and social resources. It is context-dependent, learned, and open to revision.
Self-Reflection Tool
Which relational patterns feel most familiar to you?
This brief reflection is based on the proposed Four-Factor Model of Safety Appraisal within RSF. There are no right or wrong answers, only an invitation to notice what resonates.
Anonymous · ~3 minutes · 12 questions
A gentle note: This is a conceptual self-reflection tool, not a validated measure or clinical assessment. If what you notice feels significant or distressing, please consider reaching out to a mental health professional.
How Patterns Form
Your predictions aren't broken — they were learned
When you grow up in an environment where rejection, unpredictability, or emotional unavailability is common, your brain learns to expect it. These aren't irrational thoughts — they're survival-based predictions shaped by real experience.
A child who was repeatedly dismissed doesn't just feel sad. Over time, they come to predict: "My needs won't be met," "I'm too much," "Connection isn't safe."
These predictions are not stored as a score in bodily tissue. RSF conceptualizes them as predictive relational threat structures: learned, higher-order relational models that shape attention, interpretation, expected cost, and action selection. A delayed text may be inferred as abandonment, neutral feedback as criticism, or distance as impending rejection.
Protective Strategies
Your attachment patterns make sense
When the brain infers that connection carries a high probability or cost of harm, protective responses become more likely. RSF views attachment patterns not as disorders or personality flaws, but as adaptive strategies: learned ways of acting that once reduced exposure, preserved connection, or increased control.
Some people become hypervigilant — amplifying needs, monitoring for signs of rejection, seeking constant reassurance. Others learn to shut down — minimizing needs, suppressing emotions, maintaining independence at all costs. Still others oscillate between approaching and withdrawing, caught between the need for connection and the fear of it.
These strategies were shaped by relational experience. The goal of therapy is not to eliminate them, but to expand the range of available responses and help learned models update when present evidence supports a different conclusion.
Relational Disconfirmation
How change actually happens
Insight can help identify the model, but durable change usually requires new relational evidence that is consistent, credible, and relevant enough to compete with what was learned before.
An old model shapes the inference
A current relational cue is interpreted through prior learning, increasing the predicted probability of rejection, criticism, abandonment, or another costly outcome.
The outcome violates the prediction
Instead of rejection, there is responsiveness. Instead of criticism, there is curiosity. Instead of abandonment, there is repair.
Prediction error becomes credible evidence
One surprising outcome may be treated as an exception. Repeated, consistent, context-relevant evidence gives the brain more reason to revise the old estimate.
The model becomes more flexible
New learning does not erase the past. It expands the range of plausible outcomes and available actions, allowing the response to become more sensitive to the present context.
RSF calls this process relational disconfirmation. The therapeutic relationship can become a live source of evidence, not only a setting in which techniques are delivered. Consistent responsiveness, accountability, boundaries, and repair can challenge an old prediction and support model revision. This is learning through inference and experience, not the release of trauma stored in the body.
In Practice
What this means for therapy
Beyond symptoms
Instead of treating anxiety, depression, or relationship difficulties as isolated problems, RSF examines the learned inferences, predicted costs, and protective actions that may organize them.
Your responses make sense
Protective strategies are understood as learned adaptations, not defects to remove. The goal is greater flexibility and a wider range of possible actions.
The relationship is the intervention
Therapy provides a live relational environment where old predictions can become visible and current evidence can be noticed, tested, and interpreted differently.
Change through experience
Repeated relational evidence can help the brain revise an old model when the evidence is credible, context-relevant, and strong enough to matter.
About the Author
Stephanie Underwood
MSW, RSW · Faculty of Social Work, University of Calgary
Stephanie is a registered social worker in private practice specializing in relational trauma, attachment, and safety-based therapeutic approaches. Her work bridges clinical practice with theoretical development, grounded in the belief that understanding the mechanisms of relational distress leads to more effective, compassionate care.
The Relational Safety Framework emerged from years of clinical work with clients navigating the effects of relational trauma — and from recognizing that existing models didn't fully capture what she was seeing in the therapy room. RSF integrates an account of human relational dependency, predictive processing, attachment theory, and learning-based perspectives into a coherent account of how relational trauma develops, persists, and changes.
Stephanie incorporates RSF into her practice with clients, using the framework to guide clinical formulation and create the relational conditions in which learned predictions, threat inferences, and protective strategies can become more flexible.
Get in Touch
Interested in learning more?
Whether you're a clinician interested in the framework, a researcher exploring safety-based models, or someone seeking support — I'd love to hear from you.