Working Paper · March 2026
An empirically grounded framework integrating predictive processing, constructive-developmental theory, and the theory of positive disintegration
The computational foundation of identity
Human identity functions as a predictive compression — a generative model that trades accuracy for efficiency to enable coherent action under uncertainty. This claim draws on Karl Friston's free energy principle: all self-organising biological systems maintain themselves by minimising the divergence between their internal model and incoming sensory evidence.
The self-model — beliefs, narratives, values, expectations — is itself a generative model. It predicts how you will feel, what others will do, which actions will succeed. When predictions match experience, the system maintains homeostasis. When they don't, the system registers prediction error.
Why identity must simplify reality
The compression serves three functions: energy efficiency (simplified models reduce metabolic cost — the brain uses ~20% of resting energy despite being ~2% of body mass), social coordination (shared narratives make individuals mutually predictable), and emotional regulation (stable identity reduces uncertainty and anxiety).
But compressed models accumulate structural inaccuracies. A self-model adequate at twenty may generate systematic prediction errors at thirty-five — not because the world became unpredictable, but because the model's assumptions no longer match the person's actual situation. This is Kegan's "in over our heads" — when environmental demands exceed the complexity of the meaning-making system.
Not all prediction error is suffering
The free energy principle distinguishes epistemic value (expected information gain) from extrinsic value (preferred outcomes). When you actively seek uncertainty — exploring, learning, experimenting — the resulting prediction errors are expected at the policy level. Your model predicted it would encounter the unknown. Free energy stays low.
Suffering arises from prediction error at levels of the generative hierarchy where the model has structural dependency — where predictions are load-bearing for identity coherence. Grief illustrates this: you can fully expect a death and still be devastated, because the event-level prediction is only one among thousands that assume the person's continued presence.
Interactive: Structural Dependency Distribution
Neediness is the concentration of structural dependencies on a narrow set of sources. Non-neediness is their distribution — no single prediction's failure is load-bearing. Integration is the systematic conversion of concentrated dependencies into distributed ones.
Free Energy Response After Loss
Comparison of recovery trajectories. Orange dashed line = window of tolerance threshold.
Three ways to handle identity tension
When prediction error threatens the self-model, the system has three regulatory responses — each rational under different conditions.
Interactive: Regulatory Strategies
Signal suppression reduces the effective precision of the error signal — avoidance, distraction, numbing. Short-term stability, long-term inaccuracy accumulation. Narrative reassignment reinterprets the cause externally — blaming, ideologising, moralising. Preserves coherence but reduces learning. Model updating revises the identity structure itself — the most expensive but the only strategy that increases long-term accuracy.
Why destabilisation leads to growth or collapse
Model updating requires three conditions simultaneously. Without any one, destabilisation leads to collapse rather than development.
Nervous system can sustain high prediction error within the window of tolerance. Prefrontal cortex maintains regulatory influence over limbic activation.
External co-regulation supplements internal capacity — the "holding environment." A collaborative Markov blanket where the supporter's regulation helps maintain the window.
Capacity to hold multiple competing models simultaneously. Kegan's fourth-order consciousness — taking one's own meaning-making system as object.
Trauma, in this framework, is identity destabilisation without adequate regulation and containment. The prediction errors exceed the system's processing capacity, producing fragmentation rather than reorganisation.
Processing suppressed error signals
Shadow material is information the system has actively excluded because incorporating it would destabilise the current identity structure. Unlike ordinary ignorance, shadow content generates ongoing metabolic cost — the suppression must be actively maintained against recurring prediction errors.
Integration of shadow material reduces unconscious reactivity (because triggering information is now part of the model rather than a surprise), increases behavioural flexibility, and decreases defensive reactions.
Ethically neutral adaptive capacity
As identity integrates previously excluded information, its structural complexity increases. Greater complexity produces tolerance for ambiguity, multi-perspective reasoning, and an expanded behavioural repertoire.
Critical claim: increased identity complexity is ethically neutral. It is an increase in adaptive capacity, not moral worth. A highly integrated person can deploy that capacity in service of any goal — truth-seeking, status acquisition, service, or exploitation.
Development is not linear
Identity systems oscillate between stability phases (efficient, predictable, low cognitive load) and integration phases (destabilised, restructuring, high metabolic cost). Healthy development requires both. Excessive stability leads to rigidity; excessive integration leads to paralysis.
Developmental Oscillation
Identity complexity over time. Red dips = integration phases (destabilisation → reorganisation).
Macro-level disintegrations occur a few times per lifetime, often precipitated by major transitions — career change, relationship dissolution, bereavement, parenthood. The restructuring cycle typically spans two to five years.
Integration amplifies capacity, not direction
Identity operates within a chosen motivational framework — fitness, happiness, status, truth, service. Integration increases capacity within a game but does not determine which game is chosen. Identity complexity is a domain-general amplifier, not a moral compass.
The central insight of the framework
The free energy principle identifies two formally equivalent strategies for closing the gap between model and world: update the model to match the world (inward), or update the world to match the model (outward). Both reduce prediction error. Both are equally fundamental.
Interactive: The Two Directions
Update model to match world
Integration, shadow work, therapy, stage transitions, belief revision
Update world to match model
Self-expression, persuasion, community building, cultural production, niche construction
Click a direction to compare strategies
Integration is what happens when outward strategies fail. When reality insists on contradicting the model and no amount of niche construction will resolve the prediction error, the system faces a forced choice: chronic suppression or model updating.
This also explains resistance to others' growth: if you've built your social niche around shared predictions, someone else's integration generates prediction error for you. Your growth destabilises my outward strategy.
Context-dependent, not universally accessible
Integration is constrained by nervous system regulation capacity, developmental history, trauma exposure, social environment, and economic stability. Individuals operating under chronic material threat may rationally prioritise signal suppression: survival demands immediate predictive efficiency, not long-term model accuracy.
Signal suppression and narrative reassignment are rational strategies under constraint. They become problematic only when sustained beyond the conditions that make them necessary. The framework explicitly rejects the implication that individuals who do not develop to higher complexity have somehow failed.
Identity is a predictive compression — a generative model that trades accuracy for efficiency.
Prediction error is the developmental signal — but only structurally dependent errors produce suffering.
Free energy minimisation runs in two directions — inward (integration) and outward (niche construction). Outward is cheaper.
Integration is what happens when outward strategies fail — when reality insists on contradicting the model.
Model updating requires biological regulation, relational support, and cognitive complexity.
Increased adaptive complexity is ethically neutral — it expands capacity without determining direction.