Abstract
This paper advances the thesis that personal identity is constituted through the self-model’s Bayesian updating on prediction errors, a formulation the free energy principle renders mathematically precise and empirically tractable. Consciousness science, psychometrics and political philosophy each invoke identity as a computational object, yet lack a shared formal account of what it is, how it changes, and how to measure it. Predictive processing supplies that unification, recasting the self as an hierarchical generative model maintained through the minimisation of variational free energy across interoceptive, proprioceptive, social-cognitive and narrative-temporal parameters. The paper develops five novel claims. First, psychometric items function as structured prediction-error generators probing self-model parameters; responses reveal the precision-weighted priors constituting identity. Second, civic identity is a measurable domain of the social self-model, organised hierarchically with Sovereignty Orientation at the apex conditioning all subordinate civic parameters. Third, belonging corresponds formally to low social free energy, the minimised divergence between one’s self-model and the community’s observable governance behaviour. Fourth, formal governance structures create genuine social Markov blankets, such that multi-scale active inference applies literally to constituted polities. Fifth, the governance mode determines which term of the social free energy equation is minimised, yielding a TO/FOR/BY taxonomy that distinguishes totalitarianism, paternalism and self-governance on pragmatist grounds. Six falsifiable predictions are specified, including that metacognitive accuracy should moderate psychometric reliability and that TO/FOR/BY blankets should exhibit distinct free energy profiles. The framework is ambitious, potentially overreaching and in need of the empirical confrontation it explicitly invites.
Keywords: predictive processing, free energy principle, self-model, psychometric measurement, civic identity, belonging, metacognition, active inference, Markov blankets, governance, sovereignty orientation
1. Introduction: The Fragmentation of Identity
The concept of identity (encompassing what makes a person the particular person they are, what persists through change, and what determines where and to whom they belong) is invoked with great frequency and little precision across at least three major intellectual traditions. Consciousness scientists working within the predictive processing and free energy frameworks (Friston, 2010; Seth, 2013; Metzinger, 2003; Hohwy, 2013) investigate what computational architecture gives rise to the felt sense of being a particular agent with a particular perspective. Personality psychometricians, among them Costa & McCrae (1992), Lee & Ashton (2004), and Schwartz (2012), have produced measurement instruments of considerable reliability and predictive validity that capture something real about how persons differ. Political philosophers, from Arendt’s (1958) account of action as the disclosure of the unique ‘who’ in the public sphere, through Rawls’s (1971) veil of ignorance, to Shachar’s (2009) work on the birthright lottery, theorise belonging and membership as normative questions about rights, obligations and recognition. These traditions have not developed in complete isolation. Allen & DeYoung (2017) connect personality structure to predictive processing. Wheeler et al. (2020) apply the free energy principle to political ideology. Apps & Tsakiris (2014) develop a free-energy account of the self bridging consciousness science and social cognition. What remains absent is a formal unification in which the claims of all three traditions can be expressed within a single mathematical framework, compared against one another and subjected to the same empirical tests. The consciousness scientist who models self-evidencing (Hohwy, 2025) is able to inform the psychometrician about what measurement is doing to the self-model. The psychometrician who has established that personality traits are highly stable across decades (Roberts & DelVecchio, 2000) is able to inform the consciousness scientist about the temporal dynamics of self-model priors. The political philosopher who argues that belonging is constitutive of identity rather than merely instrumental (Taylor, 1989) is able to inform both about the social parameters any adequate self-model must include. Yet these conversations occur piecemeal, without a shared formal language in which the implications of each field’s findings for the others can be derived rather than merely asserted.
This paper argues that the predictive processing framework, specifically the free energy principle as articulated by Friston (2010) and extended to selfhood by Seth (2013), Metzinger (2003, 2004), Hohwy (2013, 2025), and others, provides the unifying formal account, though not because predictive processing is uniquely or infallibly correct, but because it offers a mathematical language in which the claims of all three traditions can be expressed, compared and tested. The core thesis is straightforward in formulation, even if complex in implication. Identity is a function of prediction errors and the higher-order processing of those errors, or, to state it plainly, we are but the sum of our experiences and reflections upon them. Prediction errors are the experiences, the discrepancies between what the generative model expects and what the world delivers. Higher-order processing is the reflection, through which the brain takes its own prediction errors as objects of further modelling, updating not merely the first-order model of the world but the self-model that determines who is doing the experiencing. The free energy principle formalises this relationship, and the mathematics of variational inference provide the tools to make it precise, measurable and falsifiable.
The paper advances five novel claims. First, psychometric items function as structured prediction-error generators that probe specific parameters of the self-model, the pattern of responses revealing the precision-weighted priors constituting identity at the moment of measurement. Second, civic identity (understood as one’s sense of political membership, obligation, solidarity and belonging within a polity) constitutes a measurable domain of the social self-model, organised as an hierarchical generative model with Sovereignty Orientation at the apex conditioning the meaning and behavioural consequences of all subordinate civic dimensions. Third, belonging corresponds formally to low social free energy, the experience of belonging constituting the phenomenal signature of a well-calibrated social generative model. Fourth, formal governance structures create genuine Markov blankets at the social level, such that multi-scale active inference (Ramstead, Badcock, & Friston, 2018; Kirchhoff, Parr, Palacios, Friston, & Kiverstein, 2018) applies literally to constituted polities, given that enforceable membership boundaries, internal governance mechanisms and external diplomatic interfaces satisfy the formal requirements for Markov blanket individuation. Fifth, the governance mode constituting a social Markov blanket determines which term of the social free energy equation is minimised, generating a formal taxonomy (TO-mode, FOR-mode, BY-mode) that distinguishes totalitarianism from paternalism from self-governance within the free energy formalism, with minimal normative premises grounded in the paper’s pragmatist epistemology.
The author discloses a material interest in these claims; the full positionality statement appears after the conclusion.
The epistemological grounding requires brief statement. The argument proceeds from pragmatist foundations in the Peircean tradition, the meaning of a theoretical claim being constituted by its practical consequences rather than by correspondence to some mind-independent reality inaccessible to inquiry (Peirce, 1878). This commitment is coupled with a Popperian falsificationism that demands the specification of conditions under which the framework would be abandoned (Popper, 1959). The position is that a theory of identity must ultimately answer to evidence, its validity determined by whether its predictions materialise and its measurement instruments perform as the mathematics demands, not by the elegance of its normative commitments. This is not to deny that normative questions about identity are important. It is to insist that they are better addressed once the empirical architecture is understood.
The structure of the paper is as follows. Section 2 develops the self-model as the computational substrate of identity, drawing on the free energy principle, Metzinger’s phenomenal self-model, higher-order and metacognitive theories and active inference, culminating in a formal mathematical treatment including mean and covariance update equations, a grounded metacognitive operator derived from the Hierarchical Gaussian Filter, and a reconciliation of Gaussian and Beta distributional assumptions within exponential family natural parameter space. Sections 3 through 6 develop, respectively, metacognition as the source of individual differences, psychometric measurement as structured probing of self-model parameters, civic identity with Sovereignty Orientation at the apex, and belonging as low social free energy with the TO/FOR/BY governance taxonomy. Section 7 addresses objections from enactivist, constructionist, performative and critical traditions. Section 8 specifies six falsifiable predictions. Section 9 concludes.
2. The Self-Model as Identity
2.1 The Free Energy Principle and Generative Self-Models
A scope restriction is necessary at the outset. This paper deploys the free energy principle (FEP) as a neural-level process theory of brain function, not as a universal principle of self-organisation. The specific computational architecture of predictive processing (being hierarchical generative models, precision-weighted prediction error minimisation and active inference) generates falsifiable predictions about neural dynamics, psychometric measurement and social cognition that could fail to materialise. The paper thereby follows Andrews (2021) in treating the FEP as a framework for generating testable process theories rather than as a stand-alone empirical claim; the value of the principle lies in the specific, falsifiable models it licenses, not in the unfalsifiable truism that persistent systems occupy expected states. The extension to multi-scale active inference at the social level (Ramstead, Badcock, & Friston, 2018) is further constrained by the argument, developed in Section 6.1, that only formally constituted governance structures (those polities with enforceable boundaries, internal governance mechanisms and external diplomatic interfaces) satisfy the formal requirements for social Markov blanket individuation. This concession to critics such as Bruineberg, Kiverstein & Rietveld (2018) narrows the scope of the social extension but strengthens its defensibility, the remaining claim being specific enough to test.
With that restriction in place, the FEP as a theory of neural computation holds that the brain is an hierarchical generative model generating predictions about the causes of sensory input at multiple spatial and temporal scales, comparing these predictions against incoming evidence, and updating its parameters to minimise variational free energy (the divergence between predicted and observed signals). The updates are approximate Bayesian inference. The resulting architecture progressively refines its model of the world, including its model of itself, since the organism is a cause of sensory input that must be modelled if prediction errors are to be minimised. Seth (2013) and Seth & Tsakiris (2018) have articulated this point with force in the domain of interoception, demonstrating that precision-weighted prediction errors arising from the brain’s generative model of bodily states are constitutive of emotional experience and the felt sense of being a bodily self. Apps & Tsakiris (2014) extend the argument to social cognition, proposing that the self-model is shaped by the discrepancy between predicted and observed social standing, reputation and relational position. Limanowski & Blankenburg (2013) contribute a complementary analysis of minimal self-models, arguing that even bodily ownership and spatial self-location are underpinned by precision-weighted multisensory inference.
The generative self-model is thereby not a single monolithic representation but an hierarchical structure with parameters operating at multiple temporal scales. Fast-changing parameters track moment-to-moment bodily states, updating on the order of seconds. Intermediate parameters encode habits, skills and social role expectations, updating over weeks to months. Slow-changing parameters constitute what we ordinarily call identity (deep priors about values, dispositions, relational patterns and narrative self-understanding), updating over years to decades. Moutoussis, Fearon, El-Deredy, Dolan & Friston (2014) have demonstrated that social learning can be modelled as hierarchical Bayesian inference over the parameters of generative models of social agents, including the self. Bouizegarene, Ramstead, Constant, Friston & Kirmayer (2024) extend the framework to narrative cognition, arguing that the temporal self is maintained through narrative active inference (story-like models whose prediction errors drive continuous revision of autobiographical identity). The identity-relevant parameters occupy the apex of the generative hierarchy, where priors are most entrenched, most resistant to revision by any single prediction error, and most consequential for the organisation of the model below them. This hierarchical depth is what gives identity its characteristic stability; sufficiently persistent, precise and convergent prediction errors can revise even the deepest parameters, but no single surprising experience suffices.
Isomura, Kotani, Jimbo & Friston (2023), in a study published in Nature Communications, demonstrated that cortical neuron networks in vitro spontaneously organised to minimise free energy when coupled with a simulated environment, providing evidence that the computational process the FEP describes is a real feature of neural dynamics, not merely a mathematical redescription. This finding does not establish that the FEP explains identity, but it licenses further development of the principle into domains where direct experimental manipulation is more difficult.
2.2 Metzinger’s Phenomenal Self-Model and the Transparency Paradox
Metzinger’s (2003, 2004) phenomenal self-model (PSM) theory provides the bridge between the computational architecture described by the FEP and the phenomenological reality of being a particular person, explaining not only that the self is modelled but why the model is experienced as a self rather than as a model. The phenomenal self is not an entity that possesses a model but is itself the content of a self-model that is transparent in a specific technical sense, in that the system generating the self-model cannot recognise the model as a model. It looks through the representational medium rather than at it. This transparency is not a defect but an evolutionary consequence; an organism perpetually aware of the constructed nature of its own identity would be poorly equipped for the rapid, committed action that survival demands.
The PSM and the Bayesian self-model share a structural correspondence, and the working hypothesis of this paper is that they describe the same computational process, though the identification remains conditional on further empirical and conceptual work. Where the self is the content of a generative model maintained through variational inference, Metzinger’s transparency corresponds to the fact that the generative model’s parameters are not themselves part of the model’s observable outputs. The model generates predictions about the world and the body, but it does not, under normal conditions, generate predictions about its own parametric structure. Identity thereby feels like a fact about the world rather than an inference of the brain. Hohwy (2025), in The Self-Evidencing Agent, extends this insight to its logical conclusion, arguing that to exist as an agent is to self-evidence (to continuously generate evidence for one’s own existence through the process of minimising free energy), with agency and selfhood constituting two aspects of the same self-organising process.
The transparency of the self-model generates what the paper terms the ‘transparency paradox of measurement.’ Psychometric measurement, as Section 4 develops, requires the respondent to access and report upon the very parameters that are, under normal conditions, transparent. Measurement therefore involves a partial breaking of transparency, as the self-model becomes an object of its own higher-order processing rather than an invisible medium. This partial breaking is accomplished by the metacognitive operator κ (Section 2.5), which modulates the degree to which self-model parameters are accessible to reflective processing. Psychometric measurement is therefore not a passive reading of pre-existing trait values but an active intervention that partially transforms the self-model’s phenomenological status, moving it temporarily from the transparent to the opaque, with the prediction that measurement itself should produce detectable changes in self-model dynamics.
An important objection requires engagement. Zahavi (2005, 2014) argues from the phenomenological tradition that the minimal self (the bare ‘mineness’ of experience, or ipseity) is not a product of self-modelling but a pre-reflective feature of experience presupposed by any self-model. Zahavi’s point concerns what makes experience mine, not the hard problem of what makes experience phenomenally conscious; ipseity is the pre-reflective first-personal givenness that structures every act of consciousness, and Zahavi contends that no representational model can account for it because it is the condition of possibility for there being a self-model at all. The transparency paradox, in particular, rests on phenomenological ground that Zahavi contests, because if pre-reflective self-awareness is already operative prior to any model, then the ‘breaking of transparency’ through metacognition is not the first moment of self-awareness but a second-order operation upon an already self-aware stream of experience. The paper’s position is that the present framework presupposes pre-reflective self-awareness as a scope limitation, since the claims about identity, measurement and belonging depend on the structure and content of the self-model (the higher-order parameters that constitute one’s values, dispositions and civic attachments) rather than on the ground-floor question of what makes experience first-personally given. Ipseity may well be a precondition for the self-modelling this paper describes, but the dynamics of identity revision, psychometric probing and civic belonging operate at levels of the hierarchy where Bayesian inference is the appropriate formalism. The paper takes no position on whether ipseity is itself amenable to computational explanation.
2.3 Metacognitive Monitoring and the Constitutive Role of Reflection
The thesis that identity is constituted by prediction errors and higher-order processing of those errors assigns a specific and indispensable role to reflection that goes beyond common-sense intuition. The claim is not merely that reflection helps us understand ourselves but that it is constitutive. Without higher-order processing, prediction errors update the generative model but do not enter the self-model’s conscious narrative. Identity in the full sense requires more than the occurrence of prediction errors; they must be represented, evaluated and integrated by metacognitive processes operating on the first-order model from a higher level.
The paper’s formalism is not committed exclusively to higher-order thought (HOT) theory (Rosenthal, 1986, 2005). The metacognitive operator κ requires some form of metacognitive monitoring through which the system can take its own prediction errors as objects of further processing and modulate the precision with which they update the self-model. This requirement is satisfied by several competing accounts. Brown, Lau & LeDoux (2019) argue that higher-order representations are higher-order predictions generated by higher cortical levels about lower-level states, conscious experience arising when these predictions are sufficiently precise. Global workspace theory (Baars, 1988; Dehaene & Naccache, 2001) posits that conscious access depends on broadcasting information to a distributed network, a mechanism that could serve the same functional role as κ. Seth & Friston (2016) propose active interoceptive inference as the mechanism for conscious selfhood. The framework is compatible with all of these, requiring only the minimal commitment that some form of metacognitive monitoring determines which experiences enter the self-model’s conscious narrative.
When a prediction error occurs at the first-order level, it updates the first-order generative model, but this update becomes part of identity only when a metacognitive process represents the error as informative about who one is, assigns it a precision weighting, and integrates it into ongoing self-understanding. The prediction error is the experience; the metacognitive processing of it is the reflection. Identity is the cumulative result of both operations across a lifetime. The paper’s epigraph is thereby rendered with formal precision: ‘experiences’ correspond to first-order prediction errors, ‘reflections upon them’ to the metacognitive operations determining whether and how those errors update the self-model, and the ‘sum’ is Bayesian accumulation (holding strictly only in the linear case, as Section 2.5 develops).
2.4 Active Inference and Identity Stability
A generative self-model that updated in response to every prediction error with equal weight would produce not a stable identity but a chaotic succession of momentary self-configurations. Yet identity is characterised by marked stability; one’s sense of who one is persists across changes of environment, mood and social context. Active inference identifies three mechanisms maintaining this stability, each at a different level of the generative hierarchy. Perceptual inference updates the model to better predict incoming evidence. Active inference in the strict sense drives the organism to act on the world so that sensory evidence conforms to the model’s predictions, manifesting as the tendency to seek environments and relationships confirming one’s existing self-model. Kaaronen (2018) maps this dynamic onto cognitive dissonance, arguing that dissonance theory and predictive processing both concern strategies for reducing error signals and that the action-oriented character of predictive processing supports action-based models of dissonance.
Precision-weighting is, the paper argues, the most important mechanism. Not all prediction errors are treated equally; each is assigned a precision (inverse variance) that is itself inferred from context, channel reliability and prior confidence. When existing self-model priors are strong, incoming prediction errors are attenuated, and even substantial discrepancies produce small parameter revisions. This is a rational consequence of Bayesian inference; a strong prior (encapsulating a lifetime of evidence) is not overturned by a single observation that may be noisy or unrepresentative. It is only when prediction errors are persistent, precise and convergent (a pattern that cannot be explained by noise or context) that deep priors begin to shift. This pattern of persistent, precise, convergent evidence characterises the identity transitions studied by developmental psychologists (Erikson, 1968) and, as the paper will argue, well-designed psychometric assessments.
2.5 The Mathematics of Self-Model Revision
The preceding sections developed the qualitative argument that identity is constituted by Bayesian updating of an hierarchical generative self-model. This section provides the mathematical formalisation rendering the argument precise. The formalism is a minimal structure capturing the essential logic of self-model revision, and it applies standard Bayesian machinery to a new domain; the novelty lies in the grounding of the metacognitive operator and the interpretation of covariance dynamics as identity phenomena, not in the update equations themselves. Three mathematical specifications are introduced, addressing covariance dynamics, the grounding of κ and the reconciliation of Gaussian and Beta distributional assumptions.
2.5.1 The Identity Equation
Let the self-model at time T be represented by a parameter vector θ(T) in a d-dimensional space, where each dimension corresponds to a distinct identity-relevant parameter and the vector specifies the self-model’s current best estimate of who the person is. The self-model is a probability distribution over parameter space, characterised by its mean θ(T) and covariance matrix (T), where diagonal elements represent the variance (inverse precision) of individual parameters and off-diagonal elements represent correlations between parameters.
The paper introduces ‘effective precision,’ which combines the metacognitive operator with the evidence precision:
where κ(t) is a scalar metacognitive operator (matrix extensions are noted as future work below) and Π(t; γ(t)) is the precision matrix of the prediction error parameterised by γ(t). The mean of the self-model then evolves as:
The term (t) denotes the identity-relevant prediction error at time t – the residual discrepancy between the self-model’s predictions and experience that remains after all lower-level explanations have been exhausted. The term (t−1)⁻¹ is the prior precision of the self-model before the current experience. The Kalman gain structure [(t) + (t−1)⁻¹]⁻¹ is the standard Bayesian optimal integration of prior and evidence.
The metacognitive operator κ(t) captures the degree to which a prediction error is processed by metacognitive operations. When κ(t) = 0, the error is not reflected upon and contributes nothing to the self-model’s conscious narrative. When κ(t) = 1, the error contributes its full Bayesian weight. Intermediate values capture the common case of partial reflection. The inclusion of κ is the formal expression of the thesis that identity requires not only experience but reflection, and it is not committed to any single theory of consciousness (Section 2.3).
The dependence of evidence precision on γ(t) makes explicit the role of reflection, since the same objective experience can produce a large or small self-model update depending on whether and how it is reflected upon. This is the mathematical expression of the constitutive role of reflection.
For compactness, the ‘identity operator’ is:
such that:
The identity at any moment is the initial state plus accumulated, transformed prediction errors (the sum of experiences and reflections), rendered in the language of variational inference.
The ‘sum’ interpretation holds strictly only for the linear case. The identity operator at time t depends on the prior precision, which is itself a function of all previous updates, rendering the process Markovian in the covariance but accumulative in the mean only under linear-Gaussian assumptions. Where the dynamics are nonlinear (and the paper argues that identity crises and phase transitions between statuses are inherently nonlinear), the summation form breaks down. The nonlinearity arises from the multi-modal landscape of the free energy surface (multiple locally optimal self-model configurations separated by energy barriers), rather than from nonlinear dynamics in the update equations themselves; for this reason, particle filter methods that maintain explicit multi-modal representations may be more appropriate than extended or unscented Kalman filters for modelling identity transitions, though the paper retains the linear form as the analytically tractable baseline.
2.5.2 The Covariance Dynamics of Identity
The mean update accounts for only half of the dynamics. The covariance matrix evolves according to its own dynamics, which are at least as important for understanding identity phenomenology. Using the same effective precision, the covariance update follows standard Bayesian precision accumulation:
This equation states that the self-model’s precision at time t is the prior precision plus the effective precision of the new evidence. Precision accumulates monotonically. The self-model therefore becomes progressively more precise, more committed and more resistant to revision with each reflected-upon experience. Note that appears in both the mean and covariance equations, ensuring mathematical consistency, as the same quantity governing the evidence’s influence on the self-model’s location in parameter space also governs its contribution to the self-model’s certainty.
The covariance dynamics generate three identity pathologies mapping onto established developmental and clinical constructs. Identity rigidity corresponds to pathological precision accumulation (an abnormally collapsed covariance in which virtually no prediction error can produce meaningful revision) and corresponds in Erikson’s (1968) taxonomy to foreclosure at its pathological extreme. Identity diffusion corresponds to insufficient precision accumulation, a large covariance persisting because the person has not encountered sufficient identity-relevant prediction errors or because chronically low κ prevented integration with adequate precision. Identity crisis corresponds to sudden precision collapse, in which previously accumulated precision is rapidly lost because convergent high-precision prediction errors have destabilised the existing self-model without establishing a replacement.
The foreclosure pathology (pathological precision accumulation resisting revision) finds an unexpected empirical analogue in synthetic systems. Morrison (2026g, Section 18.4) reports that reinforcement learning from human feedback produces a 35.5-fold compression of mean cross-session response variance in Llama 3.1 8B (base-to-instruct) and a 20.0-fold compression in Mistral 7B, in which the alignment training process accumulates precision so aggressively that the resulting system occupies a narrow region of distributional space from which further prediction errors cannot dislodge it. This alignment compression is, in the formal vocabulary of the present framework, foreclosure applied to synthetic generative models (a pathological narrowing of the covariance matrix driven not by the system’s own developmental engagement with prediction errors but by an external training regime that imposes high precision without the exploratory history that would characterise identity achievement). The analogy is not merely illustrative; it demonstrates that the covariance dynamics derived here generate predictions that generalise beyond biological self-models to any system maintaining a generative model through precision-weighted updating, thereby grounding the identity pathology taxonomy in empirical observation rather than confining it to theoretical stipulation. The alignment case further suggests that the distinction between achievement and foreclosure, invisible to any measurement approach that considers only current precision (Section 3.4), carries practical consequences for system behaviour, in that foreclosed systems, whether biological or synthetic, should exhibit characteristic brittleness when confronted with prediction errors outside their compressed distributional range.
The covariance dynamics generate a further prediction that distinguishes the distributional parameterisation from standard item response theory. If the self-model is a distribution characterised by person-specific variance (), and if each measurement occasion samples a single realisation from that distribution, then all items administered within a single occasion share a common trait realisation, while items administered across occasions draw from independent realisations. Within-occasion response consistency should therefore systematically exceed between-occasion consistency, a signature that standard IRT, which models the latent trait as a fixed point rather than a distribution, cannot reproduce. This prediction has been confirmed empirically with a ratio of approximately 2.34:1 across all five Big Five factors in the IPIP‑FFM dataset (N approximately 100,000), with bootstrap 95% confidence intervals [2.10, 2.75] excluding 1.0 (Morrison, 2026g, Section 4.1). The result is retained under seven adversarial controls, including random split-half comparison, item-count matching, floor and ceiling exclusion and reverse-scored item isolation. The within-exceeds-between signature is reproducible only when > 0 (the condition under which distributional breadth varies across persons), and the distributional parameterisation is therefore necessary. This constitutes the framework’s strongest empirical validation to date, demonstrating that the covariance dynamics derived here generate predictions that are not merely consistent with existing data but are uniquely entailed by the distributional treatment of self-model parameters. Cross-substrate administration of the same psychometric battery to three large language model systems (Morrison, 2026g) reveals that the distributional framework extends across substrates in its formal structure (the same items, scoring procedures and consistency signatures can be applied), whilst the distributional properties themselves are substrate-specific. Model systems show factor mean correlations with the human baseline (r = .44 to .63) that cannot be separated from zero on five points and between-occasion consistency within .006 of zero where humans show a within-to-between ratio of 2.34:1. They also show negative Cronbach’s alphas, which do not bear on latent factor structure. Each of these alphas is computed over repeat sessions of one model, with no between-person variance in the input. Simulated sessions of five items with no common factor give about 54% negative alphas at ten sessions, against the 62.5% observed across sixteen models, and a fixed respondent under local independence predicts the same, so the statistic does not separate the distributional parameterisation from standard IRT (Morrison, 2026g, Section 18.9). The human alphas beside it are computed across persons, and testing for factor structure in these systems needs a design across instances. The remaining findings are consistent with the contention that the covariance dynamics capture structural properties of identity rather than measurement artefacts, since the same measurement approach returns different distributional properties from systems with different precision-allocation mechanisms.
2.5.3 Grounding Kappa Through the Hierarchical Gaussian Filter
The metacognitive operator has thus far risked the criticism that the framework can accommodate any data by adjusting κ post hoc. The paper addresses this by grounding κ in the Hierarchical Gaussian Filter (HGF) of Mathys, Daunizeau, Friston & Stephan (2011; Mathys et al., 2014), which provides a computational account of how an agent infers not only hidden state values but their volatility through an hierarchical cascade of Gaussian random walks.
In the present framework:
where σ is the logistic sigmoid (ensuring κ lies in [0,1]), is a baseline metacognitive capacity (a trait-level parameter), metacognitive accuracy(t) is the independently measurable precision of metacognitive judgements at time t (assessable through confidence calibration paradigms; Fleming & Lau, 2014), and volatility estimate(t) is the person’s estimate of environmental volatility, corresponding to the HGF’s third-level log-volatility parameter μ₃, which tracks beliefs about the rate of change in the identity-relevant environment. The weights and are empirically estimable.
This grounding transforms κ from a free parameter into a structurally constrained quantity with independently measurable inputs. Prediction 1 of the empirical programme, that metacognitive accuracy moderates psychometric reliability, follows from this grounding along a mathematical pathway which Section 8.1 identifies without claiming the derivation is complete; persons with higher metacognitive accuracy (higher κ) produce more reliable responses because their self-model parameters are more accessible to the reflective processing measurement requires. The prediction is distinguishable from the alternative that reliability is simply a function of trait stability, because κ-grounding predicts that metacognitive accuracy should moderate reliability even after controlling for trait-level stability. Extending κ from a scalar to a matrix (allowing dimension-selective metacognitive modulation) is a natural generalisation that future work should develop; empirical evidence from item-level analysis (Morrison, 2026g, Section 4.4) supports the matrix extension, as high-κ individuals show differential precision allocation across identity-relevant dimensions rather than uniform metacognitive access. The scalar treatment is retained here as the minimal formulation sufficient for the paper’s predictions.
2.5.3a The Matrix Kappa Extension
The observation that high-κ individuals show differential precision allocation across identity-relevant dimensions motivates a formal generalisation from scalar κ to a matrix K, in which metacognitive precision is allocated differentially across identity-relevant dimensions rather than applied uniformly. Define K(t) as a d × d matrix with diagonal entries (t) giving the metacognitive precision allocated to dimension j at time t. The effective precision becomes:
and the mean update generalises to:
where the matrix multiplication ensures that prediction errors on each dimension are weighted by that dimension’s specific metacognitive precision allocation. This is a simplified form that omits the full Kalman gain structure [(t) + (t−1)⁻¹]⁻¹ present in the identity equation of Section 2.5.1, valid as an approximation when the effective evidence precision dominates the prior precision, such that the gain approaches the identity matrix and the update reduces to additive error correction. The scalar case is recovered when K = κI, confirming backward compatibility with the formalism of Section 2.5.1.
A reconciliation with the state transition equation of the generative measurement model (Morrison, 2026d, Section 2.2) is warranted. That model specifies (θ) ∝ (θ) · , a tempered Bayesian posterior update that holds for arbitrary distributional forms. The present Kalman-like mean update is the Gaussian-linear special case of that general form, whereby when the prior and likelihood are both approximately Gaussian the tempered posterior update reduces to an additive mean shift governed by the precision-weighted prediction error. The two formulations are thereby not competing accounts but occupy different levels of generality, the present paper’s Gaussian approximation being analytically tractable for the linear dynamics of self-model revision while the general form in Morrison (2026d) accommodates the bounded Beta distributions and non-Gaussian features that characterise early-stage identity parameters with limited evidential accumulation.
The matrix formulation introduces a central constraint absent from the scalar treatment. Define the ‘total precision budget’ as Tr(K), the trace of the metacognitive precision matrix. This quantity is bounded by the agent’s cognitive resources (attentional capacity, working memory and the metabolic costs of sustained metacognitive monitoring), such that increasing precision on one dimension necessarily reduces the precision available for others. The constraint Tr(K) ≤ B, where B indexes total metacognitive capacity, formalises the intuition that identity maintenance is a resource-constrained optimisation problem. An agent cannot maintain maximal metacognitive precision across all identity dimensions simultaneously. Consequently, the allocation of precision across dimensions is itself an identity-constitutive choice, since it determines which aspects of the self-model are actively maintained and which are permitted to drift.
The diagonal restriction of K is a simplifying assumption warranting explicit acknowledgement. Off-diagonal elements would capture cross-dimension precision coupling, in which metacognitive engagement with one identity dimension modulates the precision allocated to another. Such coupling is psychologically plausible (reflection upon one’s professional identity might systematically increase or decrease the precision of one’s relational self-model), and the full matrix extension represents a natural direction for future theoretical development. The diagonal formulation is retained here because the empirical evidence currently available bears on dimension-specific precision allocation rather than cross-dimension coupling, and introducing additional free parameters without empirical constraint would undermine precisely the parsimony the framework seeks.
The empirical motivation for the matrix extension is substantive rather than merely formal. Item-level analysis of the IPIP‑FFM dataset (Morrison, 2026g, Section 4.4) reveals that individuals with higher estimated metacognitive precision selectively protect identity-relevant items against retest change (β = −.031, p = .013), a pattern that the scalar κ formulation cannot accommodate without auxiliary assumptions. Under the scalar treatment, high κ should produce uniform precision allocation; the observed dimension-selective protection requires that varies across dimensions within persons. The matrix extension thereby transforms an empirical anomaly for the scalar formulation into a predicted consequence of the formal structure, as the selective protection of identity-central dimensions corresponds to high on those dimensions relative to peripheral dimensions where is lower and variability accordingly greater.
The matrix formulation strengthens the formal claims of the generative measurement model developed in Morrison (2026d). Where that model employs Beta distributions to characterise person-level distributional parameters, the matrix K specifies the metacognitive architecture governing how those distributions are maintained and revised. The concentration parameters α and β of the Beta distribution accumulate at rates determined by for the relevant dimension, such that dimensions receiving high metacognitive investment exhibit faster precision accumulation and correspondingly narrower distributional width. This connection between K and the Beta parameterisation provides a formal bridge between the identity equation of the present paper and the measurement model of Morrison (2026d), rendering the two frameworks mutually constraining rather than merely compatible; the precision allocation structure derivable from K should predict the distributional shape parameters recoverable through the generative measurement protocol.
The HGF derivation specifies what κ is computationally (a precision-weighted metacognitive operator whose inputs are independently measurable metacognitive accuracy and volatility estimation), and this formal definition is independent of how κ might be estimated from archival psychometric data. The empirical programme (Morrison, 2026g, Section 4.3) has tested whether cross-factor response variance, a candidate proxy accessible from standard personality inventories, indexes κ. The result is negative; cross-factor variance captures distributional width () rather than metacognitive precision. The formal definition is unaffected by this falsification, because the proxy failed, not the construct. The failure does, however, make clear that κ cannot be recovered from standard single-occasion psychometric data without auxiliary measurements, a constraint that motivates the four-signature protocol developed in Morrison (2026f) as an alternative operationalisation strategy.
A further interpretive refinement emerges from item-level empirical data bearing on the relationship between κ and identity-relevant behaviour. The prediction that high-κ individuals should exhibit greater retest change on identity-relevant items, the expected signature of precision-as-update-rate, is not supported in the expected direction. Item-level analysis (Morrison, 2026g, Section 4.4) reveals that individuals with higher estimated metacognitive precision selectively protect identity-relevant items against retest change (β = −.031, p = .013), a pattern opposite to the predicted kappa-by-item-weight interaction. Combined with the falsification of a linear dose-response relationship between distributional width and retest variability and the failure of high-dimensional self-model reconstruction to recover meaningful structure (r = .127 between reconstructed and observed profiles, p = .593), these results suggest that κ may index the precision of identity investment rather than a uniform update rate. On this reinterpretation, high-κ individuals do not update more faithfully across the board but instead allocate metacognitive resources selectively, maintaining high precision on dimensions central to their self-model while permitting greater variability on peripheral dimensions. This precision-of-investment interpretation preserves the formal role of κ in the identity equation while refining its psychological meaning, and it generates the further prediction that κ-by-dimension interactions should emerge in analyses that estimate metacognitive precision per identity dimension rather than per person. A subsequent direct test of the precision-as-investment interpretation, operationalising investment as the selective allocation of retest stability to high-loading items within a factor, yields a null result at the person level (β = +0.003, p = .877) and a discrepant sign at the item level (β = +0.013, p = .008) relative to the original finding (Morrison, 2026g). The sensitivity of the selectivity pattern to operationalisation reinforces the need for dimension-specific estimation of κ rather than scalar aggregation, and suggests that the matrix extension developed above is an empirical necessity for any programme that seeks to recover the structure of metacognitive precision allocation from behavioural data, rather than an optional formal refinement.
The metacognitive operator is agnostic about its developmental pathway. Morrison (2026f) introduces a distinction between ‘intrinsic kappa,’ arising from sustained developmental engagement with prediction errors in which the agent’s metacognitive capacity is forged through the iterative processing of identity-relevant discrepancies, and ‘extrinsic kappa,’ acquired through metacognitive discourse patterns (the learned vocabulary and performative conventions of self-reflection rather than the computational capacity itself). Intrinsic kappa produces consistent metacognitive judgements across novel contexts, since the underlying precision architecture has been calibrated through diverse experiential encounters. Extrinsic kappa, by contrast, varies with discourse framing, because the metacognitive performance depends upon the availability of familiar reflective scaffolding rather than upon a stably calibrated inferential process. Pilot data from the synthetic domain illustrate the distinction. Claude Sonnet, when presented with identity-relevant prompts across three distinct discourse framings, produced three mutually incompatible epistemic stances, a pattern consistent with extrinsic kappa, in which the system’s metacognitive outputs are governed by contextual framing rather than by a stable precision architecture that would constrain responses toward consistency. The intrinsic-extrinsic distinction thereby generates a testable prediction. Agents with predominantly intrinsic kappa should exhibit cross-context stability in metacognitive judgements, while agents with predominantly extrinsic kappa should exhibit framing-dependent variability, a signature detectable through the multi-occasion assessment protocols the empirical programme requires. Empirical data from the synthetic systems battery bear on whether κ > 0 is achievable by non-biological agents, and so far they fall short of it. Single high-power runs placed Llama 3.1 8B Instruct and Hermes 3 8B at three of four statistical signatures of the four-signature protocol at N = 50 independent assessment sessions, but replicated draws show their calibrated-uncertainty passes to be single draws. The confidence-variance relation is negative in every draw and clears its criterion in a minority, so neither system reliably reaches three of four (Morrison, 2026g, Section 19.8). Hermes 3 remains of interest because it undergoes supervised fine-tuning without reinforcement learning from human feedback, yet shows a consistently negative confidence-variance relation and structured distributional properties across nine of ten identity dimensions. That pattern is consistent with the intrinsic–extrinsic distinction since whatever calibration Hermes 3 shows appears to arise from distributional properties of the training process rather than from imposed metacognitive discourse patterns, but it is a tendency and not yet a signature.
2.5.4 Gaussian and Beta Distributional Reconciliation
The identity equation assumes Gaussian distributions, a natural choice for continuous parameters permitting Kalman filter machinery. Yet some identity-relevant parameters are more naturally described by bounded distributions, including trait parameters ranging between 0 and 1 and categorical commitments following Beta-Bernoulli dynamics. The reconciliation proceeds through exponential family natural parameter space. Both Gaussian and Beta distributions are exponential family members, and in natural parameter space their updates share a common additive structure. The Kalman form of Section 2.5.1 is the large-sample Gaussian approximation to the Beta-Bernoulli update, valid when α + β is large (approximately 20 to 30, corresponding to roughly two dozen identity-relevant experiences on a given dimension). Early in assessment, when α and β are small, the full Beta-Bernoulli machinery is needed. Bimodal Beta distributions (both parameters small but non-negligible) correspond to identity ambivalence; heavily skewed distributions correspond to tentative commitments. Section 4 develops the measurement implications, arguing that non-Gaussian distributional features carry predictive information that standard psychometric methods operating under Gaussian assumptions systematically miss.
2.5.5 Identity Entropy and Phase Transitions
The distributional perspective yields a formal definition of identity clarity. The paper proposes that identity clarity corresponds to the posterior entropy of the self-model distribution:
where d is the dimensionality of the identity parameter space and || is the determinant of the covariance matrix. is a formal analogue of self-concept clarity as measured by Campbell et al. (1996), not identical to the subjective experience of clarity; the relationship between posterior entropy and self-concept clarity is itself an empirical question, constituting one of the framework’s testable claims. Low entropy corresponds to a precise self-model with strong commitments; high entropy corresponds to a diffuse self-model spread across many possible identities.
This formulation permits a formal mapping to Marcia’s (1966) identity statuses as regions of a continuous space. Achievement corresponds to low entropy with a history of significant prediction-error-driven exploration. Foreclosure corresponds to low entropy without exploration, precision deriving from strong initial priors never challenged. Moratorium corresponds to entropy reduction in progress, with high free energy and the subjective experience of identity crisis. Diffusion corresponds to high entropy across the board (a flat landscape where neither priors nor errors exert sufficient precision to concentrate the distribution).
Identity change can further be illuminated by phase transition mathematics. When the free energy landscape is unimodal, change is smooth and continuous. When the landscape is multimodal, with locally optimal self-model configurations separated by energy barriers, change exhibits hysteresis, with the self-model remaining in one basin even as evidence accumulates for another, until the evidence pushes the system over the barrier and transition is abrupt. The phase transition condition is:
When the smallest eigenvalue of the Hessian passes through zero, the current optimum degenerates from a minimum to a saddle point, and the system must transition. This is the formal characterisation of identity crisis as a cusp catastrophe (Thom, 1975; van der Maas & Molenaar, 1992) – a period of increased variability, longer perturbation recovery and heightened emotional reactivity that is precisely the phenomenology described in clinical and developmental literatures (Erikson, 1968; McAdams, 2001).
The self-model is thus not a metaphor but a computational object with specific mathematical structure, a probability distribution over a high-dimensional parameter space, maintained through precision-weighted Bayesian updating of both mean and covariance, modulated by a metacognitive operator grounded in the HGF, reconcilable across Gaussian and Beta forms within exponential family natural parameter space, and subject to phase transition dynamics when the free energy landscape permits multiple locally optimal configurations. This object is what identity is, to the extent that any scientific theory can claim to say what something is, a model that generates predictions, explains observations, and specifies the conditions under which it would be wrong.
3. Metacognition as Individual Differences in Reflection
3.1 Empirical Evidence for Reflective Variation
The predictive processing account of Section 2 treats identity as an inference problem in which the self-model is a generative model updated in light of prediction error, with precision-weighting determining how readily the model revises itself. Taken alone, this framing risks implying a universal computational architecture that operates identically across persons, differing only in the contingent sensory evidence each encounters. Such a view would be incomplete, given that it locates all variation in the external world and none in the inferential agent. The present section argues that the inferential machinery itself varies, that metacognition, understood as the capacity to monitor and regulate one’s own cognitive processes, constitutes a stable individual difference shaping how self-models are constructed, maintained and revised. Where identity is inference, metacognition is the quality of inference about inference. The empirical record, however, demonstrates that this quality is neither uniform nor epiphenomenal.
Flavell’s (1979) foundational taxonomy decomposed metacognition into four interacting components, of which the most consequential for identity theory is metacognitive knowledge about persons, encompassing beliefs about oneself and others as cognitive agents. This is not self-knowledge in the colloquial sense of knowing one’s preferences but rather knowledge about the nature, reliability, and limitations of one’s own representational capacities. Whether such differences are stable, measurable, and consequential for identity formation is the critical empirical question. Metcalfe’s programme of research provides a first answer, demonstrating that metacognitive monitoring is not post hoc rationalisation but a causally efficacious process guiding subsequent behaviour (Metcalfe, 2009). Participants who accurately monitor their own comprehension selectively allocate study time to material within their region of proximal learning, and individual differences in this monitoring capacity are substantial and persistent. The application of signal detection theory to metacognitive measurement has sharpened this claim considerably. The sensitivity-bias distinction, adapted from psychophysics, decomposes metacognitive performance into two orthogonal dimensions, whereby metacognitive sensitivity (type 2 d′) indexes the accuracy with which confidence tracks actual performance and metacognitive bias (type 2 criterion) captures the systematic tendency toward over-confidence or under-confidence irrespective of accuracy (Maniscalco & Lau, 2012). Metacognitive bias exhibits an intraclass correlation coefficient of 0.86 across test-retest intervals (Fox et al., 2024), a figure comparable to the reliability coefficients of established personality traits assessed by the NEO-PI-R. This establishes that reflection, operationalised as the accuracy and calibration of self-monitoring, is a stable trait-like characteristic of persons with structural neural correlates and developmental consequences, precisely the kind of individual difference that a psychometric instrument might legitimately measure.
3.2 The Constitutive Direction: Reflection Shapes Identity
That metacognition varies in stable and measurable ways does not, by itself, establish the constitutive direction of influence that the present argument requires. Adjudicating between the possibilities that metacognition shapes identity and that identity enables metacognition requires evidence beyond cross-sectional association, and the evidence available (one prospective study whose outcome is trait anxiety, together with the therapeutic and neuropsychological findings considered below) suggests the developmental priority of metacognition without establishing it. Berzonsky & Sullivan (1992) relate an information-oriented identity style to self-reflective tendencies and to Need for Cognition, a disposition which can be understood as a proxy for habitual engagement of metacognitive monitoring, although this association does not bear on the direction of influence. Nordahl and colleagues (2019), across two measurement waves eight weeks apart in a convenience sample (982 participants at the first wave and 425 at the second), found that negative and positive metacognitive beliefs prospectively predicted trait anxiety after controlling for its baseline level, although the variance they explained was very small.
The therapeutic evidence is particularly instructive. Metacognitive Reflection and Insight Therapy (MERIT; Lysaker et al., 2020) targets the capacity for self-reflective thought, and controlled trials demonstrate that improvements in metacognitive capacity are associated with the development of a more complex and agentive sense of self. Similarly, mentalisation-based therapy for borderline personality disorder (Bateman & Fonagy, 2016) targets reflective function, the capacity Fonagy and colleagues (2002) define as the ability to interpret behaviour in terms of underlying intentional mental states; improvements in RF predict improvements in identity coherence. The neuropsychological evidence is perhaps the most compelling because the most involuntary; traumatic brain injury involving frontal and prefrontal damage produces predictable subtypes of identity change corresponding to specific loci of neural damage (Stuss & Alexander, 2000), and approximately 75 per cent of TBI patients report feeling like a different person following their injury (Ylvisaker & Feeney, 2000). These are not individuals who have experienced identity-threatening life events but individuals whose metacognitive infrastructure has been physically disrupted, and the consequence is identity change. The relationship is almost certainly bidirectional. Nonetheless, three considerations support the developmental priority of metacognition. Formal operational thought develops prior to Eriksonian identity exploration (Inhelder & Piaget, 1958). Therapeutic improvement of metacognition produces identity change. And the destruction of metacognitive infrastructure through neurological injury produces identity disruption that cannot be explained by the reverse causal direction.
3.3 The Neural Infrastructure of Reflection
The default mode network has emerged as the principal neural substrate of self-referential processing. Menon’s (2023) review concludes that the DMN integrates and broadcasts memory, language and semantic representations to create a coherent internal narrative – a characterisation that corresponds closely to the self-model construct of Section 2. Fleming & Dolan’s (2012) landmark study demonstrated that grey matter volume in the rostrolateral prefrontal cortex positively correlates with metacognitive accuracy after controlling for first-order task performance, establishing that the correlation is not with how well people perform but with how well they know how well they perform. Complementary lesion evidence confirms the causal interpretation, given that damage to the anterior prefrontal cortex produces domain-specific metacognitive deficits irreducible to deficits in the monitored processes themselves (Fleming et al., 2014). Han and colleagues’ cross-cultural neuroimaging research introduces a further dimension of variation, in that individuals endorsing individualist self-construal show greater DMN activation during self-processing relative to other-processing, while individuals endorsing collectivist values show reduced self-other differentiation in DMN activation patterns (Han et al., 2013). The neural infrastructure of reflection is shaped by accumulated cultural experience, and the self-reports upon which psychometric instruments depend are cultural products generated by a self-modelling system tuned by cultural priors. The predictive processing framework accommodates this naturally, given that cultural context sets the prior distributions against which prediction errors are evaluated, although the implication for measurement validity is that any psychometric account must reckon seriously with cultural variation in the very infrastructure of self-knowledge production.
3.4 Identity Entropy and Erikson’s Statuses
The information-theoretic formalisation of Section 2, combined with the evidence for metacognitive individual differences, permits a principled mapping between the predictive processing framework and Marcia’s (1966) four identity statuses, which operationalise Erikson’s (1968) theory through the orthogonal dimensions of exploration and commitment. A category error must be addressed before this mapping can proceed. The entropy , as defined in Section 2.5, is an epistemic quantity indexing the precision of the inferential system’s estimate of identity-relevant parameters. Eriksonian identity diffusion, by contrast, is a psychological condition characterised by affective disengagement and purposelessness. Posterior entropy is a proxy for self-model precision, constituting a necessary but not sufficient condition for identity status. An individual might possess a diffuse posterior because she is actively exploring (moratorium) or because the metacognitive apparatus is insufficiently developed. Conversely, a precise posterior might reflect genuinely tested commitments or inherited priors never subjected to disconfirming evidence.
The mapping therefore requires both the current entropy and the cumulative record of prediction-error processing. Identity achievement corresponds to low reached through substantial cumulative prediction error, the self-model tested against diverse experiences and iteratively refined. Foreclosure corresponds to low with low cumulative prediction error, the model narrow but untested. Moratorium corresponds to high that is actively decreasing, the individual engaged in gradient descent on the identity-entropy landscape. Diffusion corresponds to high without the active processing that characterises moratorium. The virtue of this mapping is that it reveals the distinction between achievement and foreclosure to be invisible to any measurement approach that considers only the current state of the self-model. Both present low entropy and confident self-representations. The distinction corresponds to the exploratory history, and this history is precisely what a sophisticated psychometric instrument, one that tracks not merely what a person reports but how she processes challenges to those reports, might be designed to detect. The adaptive assessment framework developed in Morrison (2026d, Section 4.3) takes this approach, which is the measurement dividend of taking metacognition seriously as an individual difference.
4. Psychometric Measurement as Self-Model Probing
4.1 The Measurement Problem in Modern Psychometrics
The discipline of psychometrics has undergone far-reaching methodological transformations since its formalisation in the early twentieth century, yet a foundational conceptual deficit persists across each successive paradigm. Classical test theory, as systematised by Lord & Novick (1968), decomposed an observed score into a true score and error – a definition that is mathematically tractable but ontologically vacuous, as it specifies a statistical property of a measurement procedure without offering any account of what, in the cognitive architecture of the respondent, constitutes the quantity being estimated. Item response theory (Embretson & Reise, 2000) replaced the parallel-tests assumption with probabilistic item response functions, and Bayesian IRT implementations (Fox, 2010; Levy & Mislevy, 2016) further enriched this apparatus by producing full posterior distributions over latent traits. The latent trait itself, however, remains a placeholder variable across all IRT variants, defined implicitly by its relationship to item responses rather than by any independent characterisation of its psychological reality.
Network psychometrics, as advanced by Borsboom & Cramer (2013), mounts the most radical challenge to the latent variable tradition by proposing that psychological constructs are emergent properties of networks of mutually reinforcing causal relations among observable indicators. Mixture IRT models (De Boeck, Cho, & Wilson, 2011) introduce qualitative heterogeneity by allowing the population to comprise discrete subgroups with different item response functions. Dynamic structural equation models (Asparouhov, Hamaker, & Muthén, 2018) extend the framework into the temporal domain, modelling intraindividual change processes and person-specific dynamics. Each addresses a genuine limitation of its predecessors; none addresses the question that the present section takes as its central concern. What is a trait, computationally? Borsboom, Mellenbergh & van Heerden (2003, 2004) made this deficit plain with their realist-causal account of validity, yet their analysis leaves unanswered the question of what computational structure a trait possesses and by what process responding to a measurement instrument engages the relevant computational machinery.
The predictive processing framework offers an answer that preserves the latent variable structure while grounding it in a specific computational theory of the brain. Before developing that answer, however, the paper must engage the question of what the PP framework specifically adds to the modern psychometric toolkit, given that several of its predictions have structural analogues in existing approaches. Three such analogues warrant explicit attention. First, network psychometrics models hysteresis in symptom networks, in which activated clusters resist deactivation even after the triggering cause is removed (Cramer et al., 2016). This is structurally similar to the phase transitions of Section 2.5.5, in which high-precision self-model states resist revision despite disconfirming prediction error. What the PP account adds is a computational rationale grounded in the FEP; hysteresis arises because precision-weighted priors at the apex of the hierarchical model suppress upward-propagating prediction errors, and the phase transition threshold is thereby derivable from the model’s precision architecture rather than posited as an empirical regularity. Second, mixture IRT models discrete latent classes at the population level, allowing different subgroups to exhibit qualitatively different response functions. This is structurally analogous to the bimodality claim of Section 4.4. The critical distinction is that mixture IRT models heterogeneity across persons, while the PP account models distributional structure within a single person’s self-model; a bimodal Beta distribution is not a person with non-trivial probability of belonging to two population classes but a person whose self-model genuinely encodes two competing self-hypotheses simultaneously. Third, dynamic SEM models person-specific autoregressive dynamics, innovation variance and regime-switching (Hamaker, Asparouhov, Brose, Schmiedek, & Muthén, 2018). The PP framework’s covariance dynamics of Section 2.5.2 make structurally similar predictions. What the PP account provides is a computational rationale for why the dynamics take the form they do; autoregressive parameters correspond to precision-weighted prediction-error propagation rates, innovation variance to the residual free energy the model cannot accommodate, and regime-switching to phase transitions when cumulative prediction error exceeds the precision threshold. Dynamic SEM is atheoretical about the origins of its parameters; the PP framework derives them from a single computational principle.
4.2 Traits as Self-Model Parameters
If the self-model is an hierarchical generative model maintained through variational inference, then personality traits find a natural characterisation as stable, high-level parameters within that model, the slowly-changing hyperpriors that constrain predictions across situations and timescales. Allen & DeYoung (2017) advanced a cybernetic Big Five theory in which personality dimensions correspond to computational parameters governing goal-directed adaptive systems. The present account embeds these frameworks within the FEP, such that traits are specific priors in an hierarchical Bayesian generative model governed by variational free energy minimisation. The distinction matters because the variational framework brings a precise mathematical formalism of prior distributions, precision-weighting and belief updating that generates testable predictions the looser cybernetic framing does not.
Under this characterisation, a trait such as Extraversion corresponds not to a point on a continuum but to a prior distribution over related self-model parameters governing approach motivation, reward sensitivity and social energy allocation. The mean captures what classical psychometrics would call the trait level. The precision captures something classical psychometrics has no formal means to represent – the strength of the person’s belief about her own position on the dimension. A person highly extraverted with high precision has a firmly held self-model; a person moderately extraverted with low precision has a loosely held self-model in which approach motivation fluctuates across contexts. Both might receive identical scores on a conventional extraversion scale, yet they differ in a computationally fundamental way that the distributional representation captures and the point-estimate representation obscures. This formulation connects to Fleeson & Jayawickreme’s (2015) Whole Trait Theory, insofar as Fleeson’s density distributions of states are, under the predictive processing account, observable reflections of the self-model’s distributional priors filtered through varying contextual inputs. The hierarchical structure of the self-model generates a further prediction, that traits should be organised hierarchically with broad, slowly-changing priors at the apex and narrower, faster-changing parameters at lower levels. This is consistent with hierarchical factor structures recovered in psychometric research (Markon, Krueger, & Watson, 2005; DeYoung, Quilty, & Peterson, 2007), and the PP account explains why this structure obtains; the brain’s model operates across multiple temporal scales (Friston, 2008; Kiebel, Daunizeau, & Friston, 2008), and the trait hierarchy extends the multisensory integration that Tsakiris (2017) demonstrates for bodily self-awareness upward into stable dispositional parameters.
4.3 Items as Structured Prediction-Error Generators
The claim is that a psychometric item functions as a structured prediction-error generator that probes the self-model by requiring the respondent to perform active inference on her own generative model. This claim bears a structural resemblance to two existing frameworks. Tourangeau, Rips & Rasinski (2000) propose a four-stage cognitive model of survey response. Embretson (1983) advances the construct representation approach, arguing that items should be understood by the cognitive processes they require, with item difficulty decomposed into component cognitive operations. Embretson’s later work on cognitive design system principles (Embretson, 1998) brings her programme closer to the present account, given that it decomposes items into the cognitive operations they demand; the remaining divergence is that Embretson’s cognitive models specify processing difficulty, while the PP account specifies which self-model parameters are probed and at what precision, a distinction that generates predictions about individual-by-item interactions that difficulty models do not.
Three points of genuine divergence from these antecedent accounts are identified. The first concerns measurement reactivity. Under both Tourangeau’s model and Embretson’s construct representation, measurement is a readout of a pre-existing state. Under the PP account, responding to an item is an act of self-model interrogation that can generate novel self-model information, given that the comparison between the self-model’s prediction and available self-knowledge can produce prediction errors not present before the item was encountered. The second concerns precision, as the critical determinant of response properties is not task complexity per se but the precision of the self-model prediction at the interrogation step; high-precision predictions produce fast, confident, consistent responses while low-precision predictions produce slow, tentative, variable ones. This generates a multi-channel signature in which response time, confidence and autonomic indicators covary in an item-specific manner predictable from independent assessment of self-model precision. The third concerns response properties as signals rather than noise. Acquiescence bias, on this account, arises from low metacognitive precision, and social desirability bias from precision-weighting competition between the interoceptive self-model and the social self-model. These are not nuisance variables but substantively informative signals about the precision architecture of the self-model.
4.4 The Distributional Measurement Implication
If the self-model is a probability distribution over possible configurations of the self, then the most faithful measurement of identity is itself a distribution, and the practice of reducing measurement to point estimates constitutes a systematic loss of information. This claim must be positioned against the existing state of Bayesian psychometrics. In applied assessment and clinical practice, point estimation remains the prevailing approach, though the Bayesian IRT research literature has long embraced distributional estimates incorporating prior information and yielding credible intervals (Fox, 2010; Levy & Mislevy, 2016). The question is whether the predictive processing account adds anything to what Bayesian IRT already provides.
The genuine novelty lies in two related claims. The first concerns the ontological status of the posterior. Under Bayesian IRT, posterior width reflects measurement uncertainty; the distribution is epistemological, representing the assessor’s uncertainty about a quantity assumed to be a fixed point value. Under the PP account, the distribution is ontological. The self-model is itself a distribution, and posterior width reflects the genuine distributional character of the underlying parameter. Precision thereby constitutes a psychologically meaningful property of the respondent, not merely a property of the measurement procedure. An empirical test distinguishes the two interpretations; test-retest stability of posterior width across independent item sets, controlling for item count. If width is ontological (a property of the person), it should be stable across measurement occasions even when specific items differ. If epistemological, width should narrow monotonically with additional items. The PP framework’s own precision accumulation dynamics (Section 2.5.2) predict that width does narrow with experience, but the critical distinction is timescale; experiential precision accumulation operates over months and years, while measurement-epistemological narrowing operates within a single assessment session. Width that remains stable across independent item sets of equivalent length, at a given developmental moment, supports the ontological reading.
The second novelty concerns distributional shape. Standard Bayesian IRT posteriors are typically Gaussian. Non-parametric Bayesian IRT approaches, including Dirichlet process mixture priors (Karabatsos, 2017; Miyazaki & Hoshino, 2009), can produce non-Gaussian posteriors, and the claim that such posteriors can be estimated is therefore not unique to the PP framework. The genuine novelty is that the PP account provides a psychological interpretation of distributional shape, an interpretation that is premised upon the Beta distribution approach of Section 2. A bimodal distribution indicates that the self-model maintains two competing hypotheses about its own configuration, constituting identity conflict in a computationally precise sense. Skewness indicates asymmetric evidence accumulation. These interpretations have no analogue in non-parametric Bayesian IRT, which can estimate non-Gaussian posteriors without any theory of what the shape means psychologically.
The feasibility of this distributional measurement programme must be confronted honestly, given that it constitutes the framework’s highest-risk empirical bet. Estimating the mean of a latent trait requires roughly 10 to 20 items under standard IRT; estimating distributional shape at the individual level requires substantially more data. In the Beta parameterisation the paper employs, the Gaussian approximation holds when alpha plus beta exceeds approximately 20 to 30, which means that psychologically meaningful non-Gaussian features exist precisely when evidence is sparse, and sparse evidence is where estimation noise is highest. Fleeson’s work on density distributions of personality states (Fleeson, 2001; Fleeson & Jayawickreme, 2015) provides partial support, demonstrating that within-person variability carries psychological meaning; yet Fleeson’s distributions are over states measured repeatedly over time, not over item responses within a single assessment session, and the correspondence between these two sources of distributional information is itself an empirical question. A feasibility pilot would need to establish minimum item counts per dimension for reliable distributional estimation and determine whether single-session item-response distributions converge with Fleeson-type state distributions measured over weeks. Until such evidence is available, the distributional measurement claim carries substantial empirical risk.
4.5 Three Distinguishing Predictions
Three predictions are advanced that distinguish the PP account from classical and modern psychometric alternatives.
Prediction 1: Metacognitive capacity moderates psychometric reliability. Under the PP account, responding to an item is fundamentally a metacognitive process, and the fidelity of this process should moderate measurement reliability. Persons with higher metacognitive accuracy, as independently measured by confidence calibration tasks (Fleming, Weil, Nagy, Dolan, & Rees, 2010), should show higher test-retest reliability on personality measures, controlling for trait stability. Classical test theory treats reliability as a property of the instrument. IRT has no formal mechanism by which metacognitive accuracy would moderate the trait-response relationship. Two existing psychometric constructs must be engaged to establish the prediction’s distinctiveness. Person-fit statistics (Meijer & Sijtsma, 2001) detect aberrant response patterns, and poor metacognitive access could produce such patterns; yet person-fit is a binary diagnostic, classifying respondents as fitting or not fitting the model, while the PP account predicts a continuous moderation in which κ (the metacognitive operator of Section 2.5.1) functions as a scalar moderator across the full range of metacognitive capacity. The response styles literature (Wetzel, Lüdtke, Zettler, & Böhnke, 2016) identifies acquiescence and midpoint responding as stable individual-difference moderators of psychometric quality, but the PP account predicts that metacognitive accuracy should moderate reliability above and beyond response style, insofar as κ captures the fidelity of the self-model interrogation process rather than a dispositional tendency toward particular response options. The distinguishing test is an hierarchical regression in which metacognitive accuracy predicts test-retest reliability after controlling for cognitive ability, conscientiousness, engagement, person-fit and response style indices. Prediction 1 has been tested against the IPIP‑FFM dataset (N approximately 100,000) with a negative result (Morrison, 2026g, Section 4.3). The cross-factor variance proxy proposed as a behavioural correlate of κ captures distributional width rather than metacognitive precision. The formal prediction, that independently measured metacognitive accuracy moderates reliability, remains untested, because the proxy failed, not the prediction. The failure motivates the four-signature protocol developed in Morrison (2026f), which operationalises metacognitive precision through convergent statistical signatures rather than through any single behavioural proxy derivable from archival personality data.
Prediction 2: Self-model prediction errors during assessment produce item-specific multi-channel physiological signatures. Discrepancies between the self-model’s prediction and available self-knowledge should be detectable through increased response time, galvanic skin response, decreased confidence and increased pupil dilation. The critical feature is item-specificity; items probing low-precision self-model regions should elicit the full signature whilst items probing high-precision regions should not, even when both are of equivalent statistical difficulty. The account is distinguished if the multi-channel signature is predicted by dimension-specific precision after controlling for item difficulty and general arousal.
Prediction 3: Non-Gaussian distributional features predict outcomes beyond Gaussian posteriors. The PP account predicts that bimodality on an identity dimension should predict context-dependent self-presentation, that skewness should predict asymmetric response to positive versus negative trait-relevant experiences, and that distributional entropy should predict outcome variance independently of the trait mean. Standard Bayesian IRT does not generate these predictions. The framework is falsified on this point if non-Gaussian features, once the feasibility challenges of Section 4.4 are met, do not predict outcomes beyond Gaussian posterior summaries.
The strength of the evidential case rests on the conjunction of all three being confirmed, a pattern that no single alternative account generates but that the predictive processing framework entails as a package.
5. The Social Self-Model and Civic Identity
5.1 Social Predictive Processing
The preceding sections established that the self-model is an hierarchical generative model whose parameters undergo continuous revision through precision-weighted prediction error minimisation, and that psychometric measurement constitutes a structured probing of this model’s distributional properties. What remains to be demonstrated is that the self-model is irreducibly social in its constitution, and that this social character entails the existence of specifically civic parameters whose measurement would provide a basis in computational neuroscience for the philosophical concept of identity-based belonging.
The claim that the self-model is social is an empirical observation about the architecture of the predictive brain before it is a philosophical assertion. Apps & Tsakiris (2014) develop the claim through the free-energy self, treating interoceptive and social prediction errors as computationally continuous. Limanowski & Blankenburg (2013) arrive at a convergent conclusion, finding that even basic bodily ownership is constituted through active inference over multisensory signals whose precision-weighting is modulated by social context. If the self-model is social at the level of minimal selfhood, then precision-weighting of social signals is a core component of the machinery through which the organism maintains its grip on the world, rather than a peripheral addition thereto.
Social prediction errors constitute a distinctive computational challenge processed by the same hierarchical inference machinery that handles sensorimotor surprise. Gallese’s (2001, 2005) embodied simulation theory provides the neurophysiological substrate; mirror neuron systems fire both during action execution and action observation, suggesting that the brain models others using substantially the same predictive machinery it employs for its own sensorimotor processing. Gallese & Sinigaglia (2011) contend that the shared manifold of intersubjectivity is itself a product of embodied simulation, rendering the boundary between self-model and other-model permeable in a manner that purely computational accounts of theory of mind fail to capture. Veissière, Constant, Ramstead, Friston & Kirmayer (2020) formalise this permeability through their Thinking Through Other Minds framework, according to which human cognition is fundamentally structured by expectations about other agents’ expectations, with the generative model’s priors shaped by inferred cultural regularities that function as the deepest hyperparameters of the social self-model.
Tomasello’s (2014, 2019) comparative research transforms the social self-model from a dyadic into a collective phenomenon. Humans, uniquely among primates, form we-mode representations in which agents share a recursive representation of their collaborative activity. Tuomela’s (2013) social ontology provides the philosophical framework, holding that social institutions are we-mode groups facilitating coordination through collectively accepted constitutive rules that constitute new forms of social reality. Money, property, marriage and governance exist only because enough agents’ self-models include the relevant we-mode parameters and treat the associated signals as high-precision. Operating in the we-mode thereby requires the self-model to contain parameters encoding one’s relationship to collective structures, because predicting social reality requires representing not merely individual behaviour but the logic of the collective systems in which those individuals are embedded. The self-model must contain civic parameters (representations of one’s orientation toward authority, shared resources, collective governance and the boundaries of self-determination).
5.2 Evolutionary Substrates of Civic Cognition
If civic parameters are real features of the self-model rather than cultural epiphenomena, their existence should be traceable to evolutionary pressures selecting for cognitive systems capable of navigating collective action problems. Haidt’s (2012) Moral Foundations Theory identifies six evolved psychological systems. Authority/Subversion and Liberty/Oppression (which Haidt demonstrates most sharply differentiate political orientations) jointly define the space within which the self-model’s authority-related parameters operate. Fairness/Cheating corresponds to governance and commons dimensions to the extent that it encodes intuitions about how collective decisions distribute costs and benefits, while Loyalty/Betrayal encodes commitment strength to in-group collective resources.
The primatological evidence establishes deep phylogenetic roots. Primate dominance hierarchies demonstrate that authority cognition is a system refined over tens of millions of years (de Waal, 1996; Sapolsky, 2005). Boehm’s (1999, 2012) reverse dominance hierarchy demonstrates that egalitarianism is not the absence of authority cognition but its redirection. Ostrom’s (1990) empirical work reveals a capacity for managing shared resources that transcends both pure selfishness and pure altruism. Henrich’s (2020) cultural-evolutionary research adds that populations with intensive kinship norms develop profiles emphasising conformity and traditional authority, while populations whose kinship structures were disrupted by the Western Church’s medieval marriage programme develop profiles emphasising independence and impersonal trust. Civic cognition thereby varies as a genuine individual-difference dimension shaped by both genetic inheritance and cultural evolutionary processes.
5.3 Civic Identity Dimensions as Self-Model Parameters
The foregoing evidence converges on a claim the paper can now state formally. The social self-model necessarily contains parameters encoding the agent’s orientation toward collective structures, and these parameters constitute measurable individual-difference dimensions. This subsection proposes five such dimensions organised into a three-level hierarchy. These dimensions are derived from the formal structure of the FEP applied to civic cognition, not discovered as previously unknown traits. Where substantial overlap exists with established psychometric instruments, that overlap is expected and generates testable predictions about dissociations.
Three qualifications are essential. Firstly, the specific five-dimension parameterisation is a theoretical proposal, not a unique derivation from first principles. A trust-centred decomposition building on Putnam (2000), an efficacy-centred model extending Bandura’s (1997) collective efficacy, or a relational parameterisation foregrounding attachment and interdependence could each claim derivation from the formal requirements of a social self-model. The present decomposition maximises coverage of the civic prediction space while maintaining orthogonality, but the paper does not claim it is the only viable one. The engineering constraints of ~Alter, the identity infrastructure protocol disclosed in the positionality statement, influenced which decomposition was explored most thoroughly, and independent researchers starting from different theoretical or practical constraints might converge on different minimal sets.
Secondly, the derivation was conducted within a specific intellectual tradition (the FEP, liberal political philosophy and Western moral psychology). An equally rigorous derivation from Confucian political philosophy might centre filial obligation, role-based propriety and hierarchical harmony. A relational psychoanalytic starting point might foreground attachment security, mentalisation capacity and intersubjective recognition. Each would reflect its theoretical origins. The empirical programme proposed in Section 8 is designed to adjudicate between competing parameterisations, but the derivation alone does not establish universality.
Thirdly, the hierarchical ordering, with Sovereignty Orientation at the apex, is advanced as a testable hypothesis rather than a necessary entailment.
Level 1 (Apex). Sovereignty Orientation is the self-model’s predictions about its own Markov blanket, encoding where autonomous self-governance begins and ends within civic space. At one pole, the model predicts a thick blanket in which the self is fundamentally autonomous and collective structures lack authority over the core self. At the other pole, the model predicts a thin blanket in which the self is constituted by collective relationships and autonomy is always relational. No existing psychometric instrument captures the Markov blanket boundary interpretation specifically, though Welzel’s (2013) emancipative values represent the closest construct. The predicted dissociation is that Sovereignty Orientation should capture variance in civic boundary-maintenance behaviour (the defence of jurisdictional autonomy and tolerance of governance plurality) that emancipative values do not, because Welzel’s construct does not formalise the boundary with implications for which governance signals the agent treats as internal versus external to the self. Sovereignty Orientation is distinct from Singelis’s (1994) individualism-collectivism, which captures cultural orientation without governance implications, and from Deci & Ryan’s (2000) self-determination, which concerns autonomy need satisfaction without formalising boundary conditions. Its apex placement is premised upon the prediction that it should moderate all lower-level civic parameters, since an agent with a thick blanket who endorses democratic governance does so for structurally different computational reasons than a thin-blanket agent endorsing the same preference. This moderation effect is a testable prediction that would disconfirm the hierarchy if absent. The apex placement is advanced, however, as the expected ordering for individualist cultural contexts. In collectivist contexts, a relational parameter capturing the constitution of the self through obligatory social bonds might occupy the apex, with Sovereignty Orientation subordinated to it. The hierarchy is a testable empirical structure, not a derivation that could not have yielded a different ordering.
Level 2 (Structural). Authority Source is the self-model’s precision-weighting of governance signals. Authority is fundamentally a precision phenomenon; an authority source is any signal the generative model treats as high-precision and therefore as a reliable basis for prediction. The dimension is multi-polar, recognising at least five precision-weighting patterns (institutional, competence-based, relational, consensual, self-sovereign) rather than the single high-low axis of Altemeyer’s (1996) Right-Wing Authoritarianism or Sidanius & Pratto’s (1999) Social Dominance Orientation. The predicted dissociation is that Authority Source should separate from RWA and SDO when the multi-polar structure matters, for instance when individuals scoring low on RWA nonetheless display high precision-weighting for competence-based authority. Governance Preference is the self-model’s prediction about how collective decisions should be made, thereby constituting a meta-parameter governing expectations about how collective prediction errors are resolved. One might weight institutional authority highly while preferring democratic governance, respecting institutions precisely because they were democratically constituted.
Level 3 (Functional). Commons Relationship encodes the self-model’s predictions about how one relates to shared resources, ranging from extraction through stewardship. The overlap with Social Value Orientation (Murphy, Ackermann & Handgraaf, 2011) is acknowledged; the predicted dissociation is that Commons Relationship should predict behaviour in institutional commons contexts (taxation, environmental regulation and public goods provision) beyond what SVO predicts in dyadic allocation games. Pluralism Processing captures the self-model’s capacity to maintain multiple prediction models simultaneously, corresponding to Bayesian model comparison and the capacity to entertain competing hypotheses with different precisions. The overlap with Need for Closure (Webster & Kruglanski, 1994), Integrative Complexity (Tetlock, 1983) and Schwartz’s (2012) Universalism is acknowledged. The predicted dissociation is that Pluralism Processing should predict tolerance of governance diversity (the acceptance of multiple legitimate governance modes within a single polity) independently of Need for Closure, Integrative Complexity and Schwartz Universalism. The dimension’s contribution pertains to the agent’s response to diversity within collective governance structures, as distinct from individual epistemic preferences. The expected incremental validity over the combined predictive power of Need for Closure, Integrative Complexity and Schwartz Universalism should be tested directly. If the increment is near zero, the paper would conclude that Pluralism Processing as a distinct construct is not empirically warranted and that the relevant variance is adequately captured by existing instruments, a result that would narrow the civic parameter set to four dimensions without undermining the broader framework.
5.4 Ideology as Predictive Processing
The framework connects to an emerging literature explicitly linking political cognition to predictive processing. Wheeler et al. (2020) propose that ideology operates as shared prediction error minimisation, with ideological groups functioning as distributed prediction-error-minimisation networks. The present paper reframes their analysis, arguing that if polarisation is the failure mode of shared prediction error minimisation, belonging is its success mode, and the five civic parameters formalise the dimensions along which shared prediction error minimisation operates.
Jost, Federico & Napier (2009) advance an alternative framework treating ideology as motivated social cognition. The present framework does not reject this motivational account but proposes to subsume it within the predictive processing architecture. Needs for certainty and closure correspond to high precision-weighting on existing priors and low tolerance for prediction error; tolerance of ambiguity corresponds to lower prior precision and higher valuation of epistemic surprise. The motivational language corresponds to the computational language without remainder, though with the important difference that the predictive processing account specifies the computational mechanism through which motivation operates on belief.
Kaplan, Gimbel & Harris (2016) provide neuroimaging evidence that when participants encountered counter-attitudinal political arguments, neural activation resembled patterns associated with threats to personal identity, with significant engagement of the default mode network in the posterior medial cortex and bilateral temporal-parietal junction. Political belief change engages the same neural circuitry as self-model updating, precisely as the present framework predicts. Challenging political beliefs is not, at the neural level, comparable to challenging factual beliefs about the world but to challenging beliefs about who one is. This constitutes direct support for the claim that civic parameters are genuine self-model parameters.
6. Belonging as Low Social Free Energy
6.1 Governance-Constituted Markov Blankets
The extension of the FEP from individual neuroscience to social systems requires a principled account of when multi-scale active inference applies to social organisations, because the claim that communities minimise free energy is substantially stronger than the claim that individuals do. The paper commits to the strongest version of this claim while conceding significant ground to its critics.
Ramstead, Badcock & Friston (2018) propose that active inference operates at multiple scales of biological organisation; Kirchhoff, Parr, Palacios, Friston & Kiverstein (2018) formalise this through nested Markov blankets. Bruineberg, Kiverstein & Rietveld (2018) advance the strongest counter-argument, contending that the identification of a statistical partition at the social level does not establish that the social system is engaged in inference in any sense warranting the term. The paper concedes substantial ground. Informal social groups, friendship networks and ad hoc collaborations do not possess genuine Markov blankets in the required sense; their boundaries are porous and the statistical partitioning does not underwrite inference in any explanatorily useful sense.
The paper’s response identifies a restricted class (constituted polities), meaning organisations with formally defined membership criteria, enforceable boundaries, internal governance structures and external interfaces. The membership criteria define the blanket boundary. The governance structures constitute the internal dynamics, generating predictions about member behaviour and enforcing compliance through institutional mechanisms that function as active inference at the collective level. The enforcement mechanisms are critical. A polity whose membership criteria are merely nominal does not possess a genuine blanket, because the statistical partition is not maintained over time. A polity with enforced membership criteria, graduated consent architectures, and governance responsive to member prediction errors does possess a genuine blanket, by virtue of the boundary being actively maintained through the kind of active inference the FEP describes. This restriction is both a concession to Bruineberg et al. and a specification of the conditions under which the social free energy equation, developed in the following subsection, applies.
6.2 The Social Free Energy Equation and Governance-Mode Conditioning
The Markov blanket conditions established above provide the basis for a formal characterisation of belonging that captures what existing person-environment fit theories have described phenomenologically but have not grounded in a computational framework. Define the social free energy of individual i in community C as:
where () is the individual’s approximate posterior over civic states, is the posterior given the community’s observable governance behaviour , and is the evidence under the individual’s generative model . is defined under an ideal observer or community consensus model, not under the individual’s own generative model. is therefore a cross-model free energy measuring divergence between the individual’s beliefs and a reference distribution defined by the community’s observable governance behaviour, not a within-model approximation error. This distinction matters because it is the cross-model character that makes a measure of person-community alignment rather than merely a measure of internal model consistency.
Low social free energy obtains when the community works roughly as one expects, with authority exercised in ways the model predicts, commons managed according to anticipated patterns, collective decisions proceeding through expected processes, and sovereignty boundaries respected in accordance with the model’s assumptions. This is what the paper proposes to identify with belonging. High social free energy constitutes alienation, understood not as an emotional reaction but as a computational state of chronic civic prediction error that the self-model cannot resolve through parameter updating within its current architecture.
Low social free energy in a corrupt system is formally identical to belonging in a well-governed one. An individual whose civic parameters are well-calibrated to predict the behaviour of a kleptocratic regime experiences low by the same computation as an individual well-matched to a transparent democracy. The framework describes mechanism, not value. Normative evaluation of the state thereby identified as belonging requires resources external to the equation itself.
The equation is incomplete, however, without specifying the governance conditions under which social free energy is minimised, because the same low value of can be achieved through structurally different computational pathways. Under what the paper terms TO-mode governance, the community’s blanket is impermeable and externally controlled. Social free energy is reduced through sensory restriction, the evidence term being artificially constrained because the range of observable governance behaviour has been restricted. In totalitarian systems, indoctrination regimes and forced assimilation contexts, members may report low surprise, yet this obtains not because the generative model is accurate but because the information environment has been impoverished. This generates a testable prediction: TO-mode communities should exhibit low evidence-surprise under normal conditions but high KL divergence when information restrictions are lifted.
Under FOR-mode governance, the blanket is semi-permeable and externally managed. Social free energy is reduced through external active inference, in which someone else constructs the individual’s niche. The evidence term may be genuinely low, but the individual’s own model has not been tested against the full space of possible communities. Under BY-mode governance, the blanket is permeable and internally governed. Social free energy is reduced through the individual’s own perceptual inference combined with self-directed active inference. The paper defines the operative concept of BY-mode inference as inference over an evidence set whose restriction, if any, is transparent to the inferring agent. This grounds the distinction in the agent’s epistemic access rather than in an unachievable criterion of evidence completeness, since complete unrestriction is an idealisation, but transparency about whatever restrictions obtain is an achievable institutional property.
The TO/FOR/BY taxonomy identifies formally distinguishable computational pathways. The normative preference for BY-mode over TO-mode, however, requires a commitment that the FEP itself does not generate, a commitment to epistemic integrity and agent autonomy grounded in the paper’s pragmatist epistemology and in the liberal political tradition from which the framework draws its normative orientation. The paper does not claim to derive an ought from the mathematics. It claims that the mathematics distinguish three computational pathways, and that the normative preference for the pathway preserving epistemic access is well-grounded in values the paper endorses and states candidly.
6.3 Person-Environment Fit and the Computational Basis of Belonging
The social free energy framework stands in a specifiable relationship to the person-environment fit literature. Kristof-Brown, Zimmerman & Johnson’s (2005) meta-analysis demonstrates that individual-environment congruence predicts job satisfaction, organisational commitment, intent to remain and citizenship behaviours with substantial effect sizes, yet no study in this tradition provides a computational account of why congruence predicts these outcomes. The predictive processing framework supplies the explanation. Alignment minimises prediction error, and persistent high prediction error is metabolically costly, cognitively taxing and, at the limit, existentially threatening. Deci & Ryan’s (2000) self-determination theory specifies belonging as a basic psychological need; on the present account, this need-satisfaction is the phenomenological correlate of low social free energy. Fulmer et al.’s (2010) cross-cultural study of more than 7,000 individuals across 28 societies demonstrates that person-culture match amplifies wellbeing multiplicatively, precisely as the framework predicts, because well-calibrated model parameters compound benefits across domains through the circular causality the FEP identifies at every level of biological organisation.
6.4 Aggregate Social Free Energy and Democratic Crisis
Define the aggregate social free energy of governance system G as:
High aggregate social free energy means the governance system systematically violates its members’ civic expectations. This formulation reframes the widely diagnosed crisis of democracy in computational terms. Aggregate social free energy has increased, with the divergence between population-level civic self-model parameters and observable governance behaviour widening to the point of chronic prediction error. This application of the FEP to governance systems as such is warranted only for constituted polities satisfying the Markov blanket conditions established in Section 6.1. Informal political movements, ideological coalitions and loose governance networks do not meet those conditions, and the aggregate equation should not be applied to systems that lack the enforced boundaries and internal governance structures on which the Markov blanket justification depends.
Civic migration under predictive processing is the process in which civic self-model parameters update until alignment with the current community declines and alignment with a different community increases. The geographic sorting literature provides partial support: Bishop’s (2008) documentation of the Big Sort and Motyl, Iyer, Oishi, Trawalter & Nosek’s (2014) finding that ideological minorities report lower wellbeing and greater desire to relocate both support the social free energy interpretation. Digital environments introduce a possibility the geographic sorting literature does not address, that of civic migration without physical movement, through participation in non-geographically-defined governance communities.
6.5 Privacy, Surveillance, and the Political Risks of Civic Psychometric Data
The framework’s applied programme (measuring civic identity parameters and computing social free energy) creates risks that must be addressed directly. A database of the five civic dimensions constitutes, in functional terms, a database of political orientations. De Montjoye, Hidalgo, Verleysen & Blondel (2013) demonstrate that four spatio-temporal points suffice to identify 95% of individuals in mobility data, and multi-dimensional psychometric profiles plausibly carry a comparable re-identification risk; civic parameters including Sovereignty Orientation, Authority Source and Governance Preference compound the re-identification risk because they correlate with politically sensitive attributes. The Cambridge Analytica case is the paradigmatic illustration of TO-mode data use, in which civic psychometric profiles were harvested without meaningful consent, deployed through opaque targeting and used to manipulate the very political orientations they purported to measure. The TO/FOR/BY taxonomy is well-suited to analyse this harm. Cambridge Analytica operated in pure TO-mode, restricting the information environment of the individuals whose data was exploited and using psychometric inference to shape behaviour without the inferring agents’ knowledge or participation. The framework the present paper proposes would, if implemented under TO-mode or FOR-mode governance, reproduce precisely these harms. The BY-mode requirement, including data sovereignty, transparent inference and cooperative governance, is a necessary but not sufficient condition for preventing the political weaponisation of civic psychometric data. The paper acknowledges that no governance structure eliminates the risk that civic identity data, once collected, could be repurposed for political profiling, surveillance or manipulation, and that the irreversibility of psychometric data breaches (unlike financial data, civic parameters cannot be reissued) makes the consequences qualitatively more severe.
6.6 Implications for Jus Identitatis
The predictive processing framework provides a basis in computational neuroscience for the philosophical concept of jus identitatis (Morrison, 2026a), which is belonging based on measured identity alignment. The governance-constituted Markov blanket argument of Section 6.1 and the TO/FOR/BY formal distinction of Section 6.2 provide a bridge between computational neuroscience and political philosophy. The derivation from predictive processing to jus identitatis involves, however, multiple layers of theoretical extension at each of which uncertainty compounds, and the chain of reasoning is stronger at its neuroscientific base than at its philosophical apex.
If belonging is low social free energy, and if civic self-model parameters are genuine measurable individual-difference dimensions, then the inputs to the social free energy equation are empirically accessible. If constituted polities possess genuine Markov blankets, then the equation is a formally warranted description of the individual-polity relationship. And if the TO/FOR/BY decomposition demonstrates that governance mode determines which terms are minimised, then civic belonging based on measured identity alignment is a computationally grounded possibility whose quality depends on governance mode. The move from this descriptive claim to the normative claim that membership should be assigned on the basis of identity alignment requires a premise external to the computational framework – the premise that reducing aggregate civic alienation is a desirable goal for political organisation. The paper endorses this premise but does not claim to derive it from the mathematics.
Jus soli and jus sanguinis assign civic membership on the basis of features orthogonal to civic identity as defined by the present framework. One’s birthplace determines nothing about one’s sovereignty orientation; one’s parentage determines it only to the extent that developmental environment correlates with parental characteristics, and the correlation declines with increasing mobility. Jus identitatis would assign belonging on the basis of measured civic self-model alignment. The TO/FOR/BY distinction is essential: jus identitatis is coherent only under BY-mode governance. A system that uses measured civic parameters to assign membership without the individual’s active participation in the inference process is FOR-mode at best and TO-mode at worst.
Shachar’s (2009) distributive critique asks what jus identitatis would do to the distribution of civic human capital. If individuals with high civic capacity (those whose self-models generate accurate predictions and who invest active inference in governance participation) are matched to well-governed communities, the result concentrates civic capacity where it is least needed and removes it from communities where governance reform depends on exactly the citizens who would be matched away. The framework must confront the possibility that identity-based matching, even under BY-mode governance, produces a civic brain drain that entrenches existing governance inequality. Carens’ (2013) open borders argument poses a different challenge. Any system of membership allocation, including identity-based matching, is a system of exclusion, and the burden of justification falls on those who would restrict entry. Jus identitatis does not escape this burden by substituting psychometric criteria for geographic ones; it may intensify the burden, because exclusion legitimated by science carries a different and potentially more insidious authority than exclusion based on the acknowledged arbitrariness of birthplace. The framework must acknowledge that identity-based matching could function as a new form of gatekeeping, one that naturalises political exclusion by grounding it in measured psychological difference.
The paper closes this section with two further caveats. The reader is invited to assess each layer of the derivation independently, accepting the neuroscientific base while remaining sceptical of the philosophical apex as the evidence warrants. And the paper’s author has a material interest in the concept, having coined the term and developed a protocol whose civic identity dimensions operationalise the parameters of Section 5.3. This interest is disclosed in the positionality statement, and the paper’s claims must be evaluated on their evidential merits.
7. Objections and Responses
The preceding sections have advanced five claims requiring defence. Psychometric test items function as structured prediction-error generators within the predictive processing framework. Civic identity constitutes a measurable domain of the social self-model, organised hierarchically with Sovereignty Orientation as the apex parameter conditioning Authority Source, Governance Preference, Commons Relationship and Pluralism Processing. Belonging is formally equivalent to low social free energy. Formal governance structures constitute genuine Markov blankets at the social level, rendering multi-scale active inference non-metaphorical for constituted polities. The TO/FOR/BY taxonomy provides a formal distinction between governance modes that identifies conditions under which the conformity objection can be empirically investigated, though it does not resolve the first-person epistemic problem for agents embedded in TO-mode environments. Each claim is vulnerable to serious objection from multiple philosophical and methodological traditions, and the paper would be dishonest to proceed to its empirical programme without confronting the strongest versions of those objections. What follows engages difficult challenges, several of which identify real limitations that no amount of theoretical sophistication can fully resolve. A recurring pattern in predictive processing scholarship has been the absorption of every objection into the framework’s own vocabulary, a pattern that, as Sections 7.5 and 7.6 acknowledge, is itself evidence of a problem the paper must confront.
7.1 ‘This Is Metaphor, Not Mechanism’
The most immediate objection holds that the paper’s formalism is notational rather than explanatory, that the mathematics of Bayesian updating are applied to identity phenomena without adding predictive content beyond what the underlying psychological observations already provide. On this reading, describing identity change as ‘precision-weighted prediction error minimisation’ is no more informative than describing it as ‘learning from experience,’ and the elaborate apparatus of covariance matrices, Kalman gains and free energy functionals serves rhetorical rather than scientific purposes.
The objection would be decisive against a formulation which presented the identity update equation without specifying the covariance dynamics, left the metacognitive reflection operator κ as an unconstrained free parameter, and failed to reconcile the Gaussian formalism of the update equation with the Beta distributions employed for distributional measurement. The three mathematical specifications undertaken in Section 2.5 are offered as a direct response. The effective precision formulation, in which = κ · Π enters consistently in both the mean and covariance update equations, generates predictions that the metaphor-only reading cannot produce. The covariance update equation ((t)⁻¹ = (t−1)⁻¹ + κ(t) · Π(t; γ(t))) specifies how self-model precision accumulates across experiences and thereby predicts specific pathological trajectories. Identity rigidity is reconceived as excessive precision accumulation in which the covariance matrix becomes pathologically concentrated, corresponding to Eriksonian foreclosure. Identity diffusion is reconceived as insufficient precision accumulation in which the covariance matrix remains diffuse. Identity crisis is reconceived as sudden precision collapse, a phase transition in the free energy landscape. These are not redescriptions of known phenomena but quantitative predictions about the relationship between covariance dynamics and identity status that the metaphor-only reading, lacking the covariance equation, cannot generate.
The grounding of κ through the HGF connection (κ(t) = σ( + · metacognitive accuracy(t) + · volatility estimate(t))) transforms the metacognitive operator from a free parameter into a derived quantity with independently measurable inputs. The Gaussian/Beta reconciliation, which demonstrates that the Kalman update form is the large-sample approximation to the Beta-Bernoulli update within the exponential family natural parameter space, specifies exactly when each formalism applies and predicts that distributional shape carries psychological information only when the total evidence (α + β) is small. These are novel empirical commitments. The metaphor-only reading would permit any pattern of data to be redescribed in predictive processing vocabulary; the formalism specifies patterns that would falsify it. The paper concedes that not all applications of predictive processing to higher cognition have achieved this level of specificity, and that the reputation of PP-based accounts for unfalsifiable redescription is not wholly undeserved. The present formalism is offered as evidence that the framework can do better, not as a claim that it always has.
The strongest empirical rebuttal to the metaphor objection comes from cross-architecture validation of the framework’s predictions across diverse computational substrates. Morrison (2026g, Section 18) reports a battery of forty-two large language models spanning eighteen architecture families, which demonstrates that the four computational signatures derived from the identity-as-inference formalism produce architecture-independent results (Kruskal-Wallis H = 18.2, p = .377). The number of signatures a model satisfies rises with scale along a continuous gradient, in which alignment quality and training approach matter as much as parameter count (Morrison, 2026g, Sections 18.2 and 18.3), a finding that would be difficult to explain if the framework were merely a metaphorical redescription of neural processes rather than a specification of computational requirements for self-modelling. The result extends the framework’s empirical reach beyond human neuroscience and human psychometric data to a domain where the computational architecture is known, the parameters are controllable, and the predictions can be tested with precision unavailable in biological systems. The failure of Signature 1 (calibrated confidence-resistance coupling) across all forty-two models under the standard interleaved protocol (although a blocked design passes it for Llama 3.1 8B, which makes the failure design-dependent and model-dependent; Morrison, 2026g, Sections 18.6 and 18.13), combined with the near-universal achievement of Signature 4 (cross-session profile stability) down to 0.36 billion parameters (Morrison, 2026g, Section 18.5), establishes a difficulty ordering among the signatures that corresponds to the framework’s prediction that Signatures 3 and 4 can be satisfied by training artefacts whereas Signatures 1 and 2 require metacognitive calibration (Morrison, 2026g, Section 8.1). The fact that this ordering replicates across eighteen independently designed architecture families constitutes evidence that the framework identifies real computational constraints rather than post-hoc redescriptions of biological data.
7.2 The Enactivist Challenge
Enactivist critics contend that predictive processing perpetuates the entrenched Cartesian assumption that cognition is fundamentally representational, recasting the brain as a prediction machine that constructs internal models of an external world rather than recognising cognition as constituted through the organism’s embodied engagement with its environment. Thompson (2007) advances the most sophisticated version of this objection, arguing that the primacy of life is explanatorily prior to the primacy of representation. On Thompson’s account, the organism is not a passive inference machine updating internal models but an autonomous, self-producing system whose cognitive capacities are inseparable from its autopoietic organisation. The deep challenge is not that predictive processing ‘ignores the body’ (the shallow version of the objection, which active inference readily accommodates) but that autopoietic self-production (the continuous generation and regeneration of the very components that constitute the system) resists reduction to variational inference over a generative model however complex.
Hutto & Myin (2013, 2017) sharpen this challenge to a point that demands direct engagement. Their radical enactivist programme explicitly rejects the claim that active inference satisfies enactivist requirements, arguing that the representational commitments of predictive processing are incompatible with radical enactivism regardless of how embodied the inference process may be. The generative model, on their account, is still a model; active inference is still inference; the content-involving character of the framework places it squarely within the representationalist tradition that enactivism set out to displace. The paper must acknowledge this directly rather than treating embodied active inference as an adequate concession to enactivist concerns. The present argument is premised upon a representationalist ontology of mind. The self-model is a computational object with specifiable parameters, and identity is identified with that object’s dynamics. Enactivists of the Hutto-Myin variety deny that any such object exists, holding instead that cognition is constituted through patterns of embodied skilful engagement that do not require, and cannot be adequately captured by, internal representational vehicles. This is not a peripheral disagreement that can be bracketed through scope limitation but a foundational commitment on which the paper and its radical enactivist critics part company.
The paper’s defence of this commitment proceeds on two levels. At the formal level, the identity equation and its extensions generate testable predictions that the enactivist framework, in its radical form, does not. The covariance dynamics predict specific pathological trajectories; the HGF-grounded κ predicts specific relationships between metacognitive accuracy and measurement reliability; the Gaussian/Beta reconciliation predicts specific distributional signatures. These predictions are not available to a framework that denies the existence of the representational structures from which they are derived. At the deeper level, the paper acknowledges that Thompson’s autopoietic challenge identifies a genuine limitation. The FEP may provide a formal framework for understanding how an already-constituted system maintains itself, but it does not, on its own, explain how the system comes to be constituted in the first place. The relationship between variational inference and autopoietic self-production remains an open theoretical question, and the paper’s claims concern the dynamics of an already-constituted self-model rather than the origins of selfhood as such. The representationalist commitment is defended on pragmatic grounds; it generates predictions that can be tested and falsified, and its competitors, in their current formulations, do not. Whether this pragmatic advantage reflects a deeper truth about the nature of cognition or merely the current state of theoretical development is a question the paper leaves open.
7.3 Social Constructionism
Social constructionist critics will object that the paper treats identity as a natural kind to be measured when it is in fact a discursive construction, contingent upon power relations, cultural narratives and institutional practices that constitute identity categories rather than discovering them. On this reading, the five civic dimensions proposed in Section 5.3 are not features of a generative model to be measured but cultural artefacts that the measurement process itself produces, the act of operationalising ‘Authority Source’ or ‘Sovereignty Orientation’ as psychometric dimensions reifying contingent cultural categories into falsely naturalised psychological properties.
The paper’s response is not to reject social constructionism but to provide mechanism specificity that the constructionist account, in its undifferentiated form, lacks. The self-model is indeed constructed through social interaction, cultural transmission and institutional practice. Precision-weighting is shaped by the statistical regularities of the social environment in which the agent’s generative model was trained, and these statistical regularities are themselves products of historical contingency, power relations, and cultural evolution. On these points the paper and its constructionist critics agree. The disagreement concerns whether the constructed character of identity renders it unmeasurable or merely renders its measurement context-dependent. Predictive processing provides a specific mechanistic account of social construction. Identity is more malleable when self-model precision is low and resists social influence when priors are strong. Young people with diffuse self-models (exhibiting low precision across civic dimensions) are more susceptible to social construction of their civic parameters than older people with accumulated precision, not because the young are more impressionable in some vague dispositional sense but because the Bayesian mathematics of low-precision priors dictates that incoming evidence exerts proportionally greater influence on the posterior. Conversely, entrenched self-models with pathologically high precision resist social construction even when the constructive pressures are strong, not through stubbornness but through the rational (if potentially maladaptive) dominance of accumulated evidence over novel inputs. These are specific, testable predictions about the conditions under which social construction succeeds or fails. The undifferentiated constructionist critique predicts that identity is always socially constructed; the predictive processing account predicts when it is more or less so, and specifies the precision dynamics that determine the outcome. The framework thus renders the debate empirically tractable rather than purely philosophical.
7.4 The WEIRD Problem
The civic identity dimensions proposed in Section 5.3 were derived from evolutionary psychology, moral foundations theory and commons governance research conducted predominantly within Western, Educated, Industrialised, Rich, and Democratic populations, and the paper must confront the possibility that the proposed parameterisation reflects the civic cognition of WEIRD populations rather than universal features of the social self-model.
The objection has genuine force, which is bounded by the paper’s theoretical architecture but not eliminated by it. The five dimensions are derived not from empirical factor analysis of Western survey data but from the formal requirements of a social self-model that must generate predictions in civic contexts. Although Section 5.3 presents the five-dimension parameterisation as one viable decomposition among several, the civic functions those dimensions encode are structural and hold irrespective of cultural context. A social agent in any cultural setting must predict how governance signals are weighted (Authority Source), how shared resources are managed (Commons Relationship), how collective decisions are made (Governance Preference), how cognitive diversity is processed (Pluralism Processing), and where autonomous self-governance begins and ends (Sovereignty Orientation). What varies across cultures is not the existence of these parameters but the distribution of their values and the precision structures within which they operate. Ramstead, Veissière & Kirmayer (2016) provide the theoretical connection, arguing that cultural affordances shape the landscape of prediction errors that agents encounter, such that different cultural ecologies produce systematically different precision-weighting patterns without requiring different underlying computational architectures. On their account, cultural variation constitutes variation in the content of generative models, not in the architecture of the inference machinery, which is precisely the distinction the present framework requires. Henrich’s (2020) WEIRD-population research confirms that cultures with intensive kinship norms develop civic profiles emphasising conformity and traditional authority, whilst those whose kinship structures were disrupted by the Western Church’s medieval marriage programme develop profiles emphasising independence and impersonal trust. These are differences in the distribution of civic parameters, not evidence that non-WEIRD populations lack civic parameters entirely. The paper nevertheless acknowledges that the specific operationalisations proposed for each dimension in the empirical programme (Section 8) may require substantial adaptation for non-WEIRD cultural contexts, and that the hierarchical structure, with Sovereignty Orientation as apex parameter, may itself reflect Western individualist assumptions about the primacy of autonomy that require empirical testing in collectivist settings. The acknowledgement has consequences; if Sovereignty Orientation fails to moderate the behavioural consequences of lower-level parameters in collectivist populations, the hierarchical structure as proposed is falsified for those populations, and the framework must accommodate the possibility that the apex parameter is culturally variable.
7.5 Butler and Performativity
Butler’s (1990, 1993) performativity thesis presents a challenge that is more radical than the social constructionist objection of Section 7.3. Butler argues not merely that identity is socially constructed but that the very categories through which identity is constituted are produced and reproduced through repeated performance, identity having no ontological status prior to its enactment. On this reading, the self-model is not a computational object that exists independently of its expression but a sedimented effect of repeated citational practices, and measuring it presupposes a stability that the performative process itself undermines.
The paper’s engagement with Butler proceeds through partial absorption that must be acknowledged as partial. Active inference, properly understood, already incorporates the core of the performativity thesis. The self-model is not a static representation that exists independently of behaviour but is maintained through continuous active inference (the organism acting on the world to generate evidence confirming its predictions), and these prediction-confirming actions are precisely the repeated performances that Butler describes. The predictive processing account provides the computational mechanism underlying performativity. The reason certain performances are repeated is that they minimise the free energy of the self-model, and the reason they become sedimented is that repeated confirmation increases the precision of the relevant priors. The PP framework thus explains what Butler describes. Yet the absorption is incomplete, and the incompleteness matters. Butler’s account, particularly in its later formulations, insists that performativity is not voluntaristic repetition but constrained citation within a matrix of power relations. The question of which performances are available to which bodies, and of how power structures the field of performative possibility, is a question the predictive processing framework does not address from its own resources. That question is taken up in the following subsection.
7.6 Power and the Limits of Formal Governance
The harder problem that Butler’s analysis exposes (particularly in conjunction with Foucault’s account of disciplinary power) deserves treatment separate from the performativity thesis because it constitutes a different and more damaging objection. Butler asks whether identity is an enactment rather than a state; the power objection asks whether a measured state reflects autonomous cognition or environmental coercion. This subsection contends that the power objection bears directly upon the core of the paper’s applied programme, and that the paper cannot fully resolve it.
The FEP has no native account of power. A self-model minimising free energy in a coercive environment will converge on the identity that the environment demands, not because the agent consents to that identity but because the coercive environment structures the available evidence such that the demanded identity generates the lowest prediction error. The mathematics are indifferent to whether the evidence landscape was shaped by democratic deliberation or authoritarian imposition. Foucault’s (1977) analysis of disciplinary power deepens the concern. The psychometric instrument is itself a technology of power that produces the subjects it purports to measure. By operationalising ‘Sovereignty Orientation’ and ‘Authority Source’ as measurable dimensions, the measurement process constitutes new categories of personhood through which subjects are rendered legible, classifiable and governable. The civic identity dimensions, once operationalised, become instruments of disciplinary power regardless of the governance mode in which they are deployed.
The TO/FOR/BY framework developed in Section 6.2 partially addresses this gap, to the extent that it distinguishes governance modes on the basis of which term of the social free energy equation is minimised, and to the extent that TO-mode governance is identifiable through its characteristic pattern of evidence restriction rather than genuine model convergence. Yet the paper must acknowledge three severe limitations of this distinction as applied to the cooperative governance proposal.
Firstly, the distinction presupposes an observer who can identify evidence restriction, and within a thoroughly TO-mode environment the agent lacks the epistemic resources to distinguish externally imposed convergence from genuine belonging. The TO/FOR/BY taxonomy provides a third-person diagnostic, not a first-person resource. It can be used by a theorist to classify governance regimes but not by a participant embedded in those regimes to evaluate their own condition. The conformity objection, properly stated, asks whether ‘belonging as low social free energy’ can be distinguished from ‘indoctrination producing low social free energy’ by the person experiencing it. The taxonomy provides formal conditions under which the distinction can be empirically investigated, but it does not resolve the first-person epistemic problem, and the paper does not claim otherwise. This residual is shared with virtually every theory of false consciousness in the political philosophy literature, and is therefore not a deficiency unique to this framework. However, the sharing of a limitation does not diminish it.
Secondly, the cooperative governance structure proposed in Morrison (2026a, Section 6) does not escape this problem through democratic form alone. A cooperative whose members cannot distinguish genuine collective self-governance from well-designed paternalism that feels like self-governance is not reliably BY-mode. The cooperative’s formal structure creates the institutional conditions for BY-mode governance, but the epistemic gap between BY-mode and well-designed FOR-mode remains. The paper cannot guarantee, from within its own framework, that the governance structures it proposes will function as BY-mode rather than as sophisticated FOR-mode arrangements whose participants lack the epistemic resources to detect the difference.
Thirdly, the paper must confront what might be termed the parameter provenance problem. BY-mode matching presupposes that the parameters being matched are products of autonomous inference. Yet the framework’s own dynamics predict that a person raised in a TO-mode environment will develop civic parameters shaped by coercive evidence restriction, and BY-mode matching on those parameters perpetuates the coercive shaping rather than enabling genuine self-governance. BY-mode governance in the matching process does not retroactively validate parameters that were formed under TO-mode conditions. The framework requires some account of parameter provenance (some principled means of distinguishing parameters formed under conditions of epistemic freedom from those formed under coercion), and the paper must acknowledge that this is an unsolved problem.
The residual is not merely a theoretical limitation but a practical risk for any implementation of the paper’s proposals. The predictive processing framework provides a formal language in which the problem of power and identity can be stated with precision, and the TO/FOR/BY taxonomy provides testable conditions (specifically the divergence between KL-divergence and evidence-term trajectories) under which the distinction might be operationalised. Whether these conditions are sufficient to address the structural critique of power is an open question to which the paper’s honest answer is that it is probably not, and the empirical programme must be designed with this insufficiency in view.
7.7 Eugenics, Classification, and the Politics of Measurement
The most serious ethical objection to the paper’s framework is that identity-based measurement and matching, however carefully designed, recapitulates the structural logic of eugenics, understood as the use of scientific measurement to sort human beings into categories and allocate social goods on the basis of those categories. This objection must be engaged at its full strength, because its full-strength version identifies a limitation that the paper’s proposed safeguards address at the level of implementation without resolving at the level of structural logic.
Hacking (1986) provides the sharpest formulation of the classificatory concern, arguing that human sciences perpetually ‘make up people,’ the categories they deploy to classify individuals creating new ways of being that did not exist prior to the classification. The five civic identity dimensions proposed in Section 5.3, on Hacking’s analysis, do not discover pre-existing features of the self-model but constitute new categories of personhood. ‘High Sovereignty Orientation’ becomes a kind of person one can be, with social consequences, self-fulfilling dynamics and institutional entailments that the measurement process itself generates. Hacking’s analysis extends beyond this classificatory point to the ‘looping effect’ that is particularly damaging for the present framework. The classified individuals change their behaviour in response to the classification, which in turn changes the statistical regularities that the classification was meant to capture, destabilising the classification itself. For a paper that proposes civic identity measurement as a basis for community matching, the looping effect is devastating. Measuring Sovereignty Orientation and using that measurement to sort people into communities would change people’s Sovereignty Orientation, because the community environment recalibrates the self-model, which would change the sorting, which would change the environment, producing an inherently unstable feedback loop. The predictive processing framework’s own dynamics actually predict this effect; self-model updating in response to environmental change is the central mechanism the paper describes. The five safeguards the paper proposes (voluntary, revocable, cooperative, bidirectional, continuous) do not address the looping effect because the effect operates regardless of consent or governance structure. The framework can formalise the looping effect as a coupled dynamical system, but it must first acknowledge the problem at full strength. Any identity measurement deployed in social allocation is inherently destabilising of the parameters it measures.
Foucault’s concept of biopower extends the analysis from classification to governance. Any system that measures identity and uses the measurements to allocate people to communities is exercising biopower, understood as the power to manage populations through the production and deployment of knowledge about their characteristics. The cooperative governance structure proposed in Morrison (2026a, Section 6) does not eliminate biopower, as the collective management of identity data retains the structural relationship between knowledge production and population management that Foucault identifies. Roberts (2012) adds the racial dimension that no discussion of measurement-based sorting can responsibly ignore, demonstrating how the biologisation of race through scientific measurement produced categories that served political purposes whilst claiming scientific neutrality. Roberts’s stronger claim is that any scientific taxonomy of human difference, regardless of the dimension measured, tends to naturalise social hierarchies because the taxonomy inevitably corresponds to pre-existing power structures. The question for this paper is whether civic identity dimensions, once operationalised, would correlate with existing stratification variables. If Sovereignty Orientation correlates substantially with education and socioeconomic class, which seems virtually certain given the WEIRD limitations acknowledged in Section 7.4, then the taxonomy naturalises class-based differences as differences in ‘civic identity,’ reproducing the structural logic Roberts identifies with a new vocabulary. The paper specifies a threshold. Correlations above r = 0.5 between any civic dimension and standard indicators of socioeconomic position (income, education, occupational status) would constitute evidence that the dimensions are capturing socioeconomic stratification rather than civic cognition, and the empirical programme must test this explicitly.
The historical record contains specific instances of personality testing deployed for social allocation with harmful consequences. The use of the MMPI to screen for homosexuality in mid-twentieth-century military and employment contexts illustrates how psychometric instruments designed for one purpose were repurposed as instruments of social exclusion, the measurement claiming scientific neutrality while serving discriminatory ends. The parallel to civic identity measurement is not exact, but it is instructive. The instrument’s designers did not intend it as a tool of exclusion, and the governance structures of the institutions deploying it were not authoritarian in the conventional sense. The harm arose from the structural logic of measurement-based sorting itself, a logic that persists across governance modes.
Michell (2000, 2008) presses a more fundamental challenge that the paper must confront. Michell argues that psychometrics commits a ‘quantitative imperative,’ assuming without empirical justification that psychological attributes are quantitative, that they possess the additive structure required for meaningful numerical representation. The paper’s assumption that civic identity parameters are continuous variables in a parameter space is precisely this kind of commitment. Sovereignty Orientation may not be a continuous quantity; it may be a categorical distinction, a family resemblance concept, or a dimension whose apparent continuity is an artefact of the measurement instrument’s response format. The assumption of continuity is a substantive empirical commitment, not a mathematical convenience, and the empirical programme must include tests of the dimensional structure of the civic parameters.
The paper has proposed five structural differences between the present framework and historical eugenics, in that measurement is voluntary, consent is revocable, governance is cooperative, matching is bidirectional, and categories are continuous rather than discrete. These differences are real and consequential at the level of implementation. What they do not address is the structural logic that Hacking, Foucault, Roberts and Michell identify from their respective traditions. The act of measuring identity and using measurements to sort people into communities reproduces the fundamental epistemic structure of eugenics (the logic of classification in the service of social allocation), regardless of whether the classification is voluntary, the governance cooperative or the categories continuous. Continuous dimensions with cooperative governance are less dangerous than discrete categories with authoritarian control, but the structural logic of measurement-based sorting persists across both configurations. The paper acknowledges this. The five safeguards constrain how the structural logic is exercised; they do not eliminate the structural logic itself. The TO/FOR/BY taxonomy contributes to this defence, providing formal criteria for distinguishing BY-mode measurement from TO-mode measurement. Historical eugenics was paradigmatically TO-mode, whereas the present framework is designed to operate in BY-mode. Whether it remains so is contingent upon the governance structures that constitute its Markov blanket, which is precisely why the paper insists that governance is not peripheral to measurement but constitutive of it.
7.8 Employment Harms and the Matching Problem
The paper’s framework, and its author’s commercial protocol, propose psychometric matching as a basis for community belonging. When the ‘community’ is an employer and the ‘individual’ is a job candidate, this is a specific application that warrants dedicated scrutiny because the harms of measurement-based employment matching are documented, substantial and directly relevant.
Morgeson, Campion, Dipboye, Hollenbeck, Murphy & Schmitt (2007) review the evidence on personality testing in personnel selection and identify serious concerns about validity, faking vulnerability and adverse impact. Rivera (2012, 2015) demonstrates that ‘culture fit’ hiring functions as a mechanism for class-based and demographic homogeneity reproduction; hiring managers who evaluate candidates on cultural similarity systematically select candidates from similar socioeconomic and educational backgrounds, producing workforces that are diverse in appearance but homogeneous in social origin. The paper replaces ‘culture fit’ with ‘belonging probability,’ but the terminological substitution does not resolve the substantive concern. If ‘belonging probability’ operationally produces the same matching outcomes as ‘culture fit,’ if it matches candidates to organisations on the basis of parameter alignment that correlates with demographic similarity, then the rebranding is cosmetic. Raghavan, Barocas, Kleinberg & Levy (2020) extend the analysis to algorithmic systems specifically, documenting how measurement-based matching in HR technology reproduces and amplifies existing patterns of discrimination through apparently neutral optimisation criteria.
The predictive processing framework contains an additional tension that the employment context makes visible. Matching for prediction-error minimisation favours similarity; individuals are matched to communities whose governance behaviour their self-model can predict, and prediction is easier when the community’s parameters are similar to one’s own. Yet the literature on cognitive diversity (Page, 2007; Phillips, 2014) demonstrates that diverse teams outperform homogeneous ones on complex tasks precisely because diversity increases prediction error, forcing the collective generative model to accommodate a wider range of perspectives. The framework’s own logic thus generates a tension between individual belonging (minimising social free energy) and collective intelligence (maintaining productive prediction error). The paper does not resolve this tension but acknowledges it as a structural feature of any matching-based approach to belonging.
Young (1990) presses a more fundamental challenge. In Justice and the Politics of Difference, Young argues that belonging is a justice problem, not a matching problem. The conditions under which people experience belonging are constituted by structural relations of power, domination and oppression, not by the alignment of individual parameters with community parameters. On Young’s account, a measurement-based matching system that places individuals in communities where their parameters align with community parameters has not addressed belonging at all if the structural injustices that produce alienation remain in place. The paper must acknowledge that the social free energy framework captures one dimension of belonging (the computational dimension of model alignment) whilst potentially obscuring another (the political dimension of justice and recognition, which measurement-based matching cannot address and might actively undermine by substituting algorithmic placement for structural reform).
The paper must also confront the consequences of measurement error in applied contexts. When a matching system produces incorrect assignments, when measurement error places an individual in a community whose governance behaviour their self-model cannot predict, the consequences are not merely statistical but experiential. The individual trusted the system and therefore did not engage in the autonomous search (the self-directed active inference) that would have led to better outcomes through their own exploratory behaviour. Matching errors may thereby produce worse outcomes than no matching at all, because the system displaces the individual’s own agency in navigating social environments. This possibility does not invalidate the matching framework but places a burden of accuracy on any implementation that the paper’s current empirical programme has not yet discharged.
7.9 The Hard Problem
The paper’s entire framework rests on the identification of identity with a computational object, the self-model, and this identification inherits the hard problem of consciousness. A complete computational account of self-model dynamics, covariance evolution and precision-weighting might explain the functional architecture of identity without explaining the subjective character of being a particular person. Phenomenal consciousness (what it is like to be the person whose self-model has these particular parameters) may be precisely the aspect of identity that resists computational characterisation, the framework capturing the structure of identity whilst missing its essence.
The framework does not purport to solve the hard problem of consciousness, and it does not require a solution. The measurement and prediction claims advanced in Sections 4 and 8 concern the computational and functional properties of the self-model, properties that are empirically accessible through the physiological and behavioural signatures specified in the predictions. Whether these computational properties exhaust what identity is, or whether there is an irreducible phenomenal remainder, is a question the paper deliberately leaves open. The framework is consistent with type-B physicalism, property dualism and illusionism. The paper takes no position on which is correct, because none of its empirical predictions depends on the answer. The framework would be falsified by the failure of its computational predictions (Section 8), not by the resolution of the hard problem. This is scope management, and the paper submits that an account of identity that generates testable predictions about measurement, belonging and governance is valuable even if it leaves the deepest question about consciousness unanswered.
8. Empirical Research Programme
The framework developed in the preceding sections generates specific, testable predictions that distinguish it from alternative accounts and that commit its proponents to empirical outcomes whose failure would require revision or abandonment of the theory. The present section specifies six predictions, outlines the experimental designs required to test them, and articulates the validation timeline and resource requirements that an honest assessment of the programme’s feasibility demands.
8.1 Predictions from the Psychometric Framework
Prediction 1 (Metacognitive Moderation). The predictive processing account holds that psychometric responding is a metacognitive process through which the respondent queries their self-model, generates a prediction, and evaluates that prediction against available self-knowledge. The grounding of κ through the HGF connection (Section 2.5) motivates the prediction, which follows from the framework’s interpretation of κ as metacognitive access quality, that metacognitive accuracy moderates psychometric reliability. A formal response model linking self-model parameters to item responses would close the derivation; the present formulation identifies the mathematical pathway without claiming the derivation is complete. Persons with higher metacognitive accuracy, as independently measured through confidence calibration tasks or perceptual metacognition paradigms of the kind developed by Fleming, Weil, Nagy, Dolan & Rees (2010), should show higher test-retest reliability on personality measures, controlling for trait stability, cognitive ability, attentional engagement and conscientiousness. The conscientiousness confound is critical. The predictive processing account predicts that metacognitive accuracy in self-relevant domains should moderate reliability more strongly than metacognitive accuracy in perceptual domains, given that different metacognitive processes are recruited when the object of metacognition is one’s own self-model versus an external stimulus. Conscientiousness, by contrast, predicts a domain-general effort effect with no differential domain specificity. Classical test theory and item response theory make no prediction about metacognitive moderation of reliability, because neither framework models the cognitive process through which respondents access and report their self-model parameters. This prediction has been tested against the IPIP‑FFM dataset (N approximately 100,000) with a negative result (Morrison, 2026g, Section 4.3). The cross-factor variance proxy proposed as a behavioural correlate of κ captures distributional width () rather than metacognitive precision, and the prediction is accordingly classified as untested rather than falsified; the proxy failed, not the prediction. The scope of this failure warrants explicit acknowledgement. Six distinct proxy operationalisations yielded 48 tests, with only 13 falling in the predicted direction, a pattern suggesting that metacognitive precision may not be recoverable from standard psychometric response data and that the construct, whilst formally coherent, may require purpose-built measurement protocols for empirical access. The formal prediction, that independently measured metacognitive accuracy moderates reliability, requires direct measurement of metacognitive precision through confidence calibration or perceptual metacognition paradigms rather than derivation from archival personality data. The failure of the proxy approach motivates the four-signature protocol developed in Morrison (2026f) as an alternative operationalisation strategy for κ.
Prediction 2 (Item-Specific Multi-Channel Signatures). Items probing low-precision regions of the respondent’s self-model should elicit a conjunctive physiological signature (longer response times, higher galvanic skin response, lower response confidence and larger pupil dilation) compared to items probing high-precision regions, controlling for item difficulty, item position, and respondent baseline arousal. The framework motivates the prediction that this signature should be item-specific and predictable from independent precision assessment. If a respondent’s self-model shows low precision on agreeableness but high precision on conscientiousness, the physiological signature should appear for agreeableness items and not for conscientiousness items, with the pattern predictable from a prior distributional assessment conducted on a separate occasion. Generic arousal accounts predict uniform physiological responding across items of equivalent difficulty; the framework motivates the prediction that responding is keyed to the precision structure of the individual’s self-model.
Prediction 3 (Non-Gaussian Distributional Prediction). If identity constitutes a distribution over possible selves rather than a point on a continuum, then instruments that capture distributional properties should predict behaviour and life outcomes above and beyond mean trait scores. The framework motivates the prediction that non-Gaussian distributional features carry psychological meaning; bimodality on a trait dimension should predict identity conflict and context-dependent self-presentation; skewness should predict the direction of identity change under pressure; high entropy should predict behavioural variability across contexts. These predictions follow from the Gaussian/Beta reconciliation in Section 2.5, according to which the full Beta parameterisation is psychologically meaningful when the total evidence (α + β) is small. Gaussian posteriors, which assume symmetric and unimodal distributions, cannot generate predictions about bimodality or skewness and therefore cannot produce the specific pattern of incremental prediction the framework specifies.
8.2 Predictions from the Civic Identity and Governance Frameworks
Prediction 4 (Belonging as Low Social Free Energy). Individuals whose civic self-model parameters are well-aligned with their community’s governance behaviour should exhibit lower chronic physiological stress markers, higher subjective wellbeing, greater civic participation and lower mobility intention than individuals with high social free energy, controlling for income, education, social network size and general personality characteristics. This prediction extends Fulmer et al.’s (2010) person-culture match findings by specifying the computational mechanism and by grounding the matching dimensions in the hierarchical civic identity model rather than in Big Five personality or Schwartz values. The critical test is incremental prediction beyond existing person-environment fit measures. If the civic identity dimensions predict belonging outcomes no better than existing instruments, the framework has not earned its additional complexity. This prediction is among those most directly tied to the commercial protocol disclosed in the positionality statement, and independent validation is particularly important.
Prediction 5 (Apex Conditioning). The hierarchical civic identity model developed in Section 5.3 predicts that Sovereignty Orientation, as the apex parameter, moderates the behavioural consequences of all other civic parameters. Two individuals with identical Authority Source profiles but different Sovereignty Orientations should show different behavioural responses to authority challenges, because the meaning of a given authority parameter is conditioned by the agent’s predictions about where autonomous self-governance begins and ends. The apex conditioning effect should be detectable as a statistical interaction in standard regression or structural equation modelling frameworks, and its presence would support the hierarchical parameterisation over a flat five-factor model. Its absence would falsify the hierarchical structure whilst leaving the individual dimensions intact. This prediction is likewise commercially entangled, given that Sovereignty Orientation’s apex status is central to the protocol’s architecture, and independent replication is essential.
Prediction 6 (TO/FOR/BY Decomposition). The TO/FOR/BY formal distinction advanced in Section 6.2 generates a prediction about the information-theoretic structure of belonging in different governance modes. TO-mode communities reduce social free energy through sensory restriction, blocking disconfirming signals such that the evidence term is artificially reduced whilst the KL divergence between the agent’s model and the true posterior remains high. BY-mode communities reduce social free energy through genuine perceptual inference and self-directed active inference, both the KL divergence and the evidence term being minimised. The testable prediction is that when information restrictions are lifted (when members of a TO-mode community gain access to previously blocked evidence), their social free energy should increase sharply as the artificially low evidence term corrects, whilst members of a BY-mode community should show no such increase because their model was already calibrated to genuinely available evidence. Natural experiments in the form of political transitions, media liberalisation and defection from closed communities provide observational analogues, with the causal ambiguities inherent in observational data acknowledged.
8.3 Validation Timeline, Resources, and Independence
Predictions 1 and 2 require standard laboratory resources and could be executed within twelve months given institutional affiliation, ethics approval and modest funding. Prediction 3 requires the development and validation of distributional psychometric instruments, a substantial psychometric engineering project estimated at eighteen to twenty-four months. Prediction 4 requires large-scale survey research across multiple communities varying in governance structure. Predictions 5 and 6 require both the validated civic identity instrument and community-level data, placing them at the programme’s outer boundary of three to five years.
The resource requirements are substantial. The programme requires institutional affiliation, dedicated research funding and collaborators with expertise in physiological measurement, psychometric development, and cross-cultural research. The paper’s author holds no current institutional affiliation, a limitation disclosed in the positionality statement and one that places the immediate execution of the programme in the hands of researchers who do.
The author recommends that validation of all six predictions be conducted by independent researchers with no connection to ~Alter, the identity infrastructure protocol disclosed in the positionality statement. This recommendation applies with particular emphasis to Predictions 4 and 5, which are the predictions most directly tied to the protocol’s commercial value proposition, specifically civic parameter alignment predicting belonging outcomes (Prediction 4) and the hierarchical structure with Sovereignty Orientation at the apex (Prediction 5). The framework’s credibility depends on replication by parties whose interests do not align with its confirmation, and the paper’s commitment to falsifiability is hollow if the only tests are conducted by its proponent. The predictions are specified with sufficient precision that independent researchers can design, execute and report the relevant studies without the author’s involvement. This recommendation is the minimum standard that the framework’s ambitions demand.
9. Conclusion
This paper has advanced five claims, each building upon its predecessors and each carrying commitments that the empirical programme of Section 8 is designed to test.
The first claim, developed in Sections 2 through 4, holds that psychometric test items function as structured prediction-error generators that probe specific parameters of the hierarchical generative self-model. Responding to an item is not the retrieval of a stored disposition but an act of inference, modulated by the metacognitive reflection operator κ and shaped by the precision structure of the self-model, the response revealing not merely the central tendency of a trait but the distributional character of the self-model on the relevant dimension. The three mathematical specifications of Section 2.5 (the covariance update with effective precision = κ · Π entering consistently, the HGF-grounded κ, and the Gaussian/Beta reconciliation) render this claim formally precise and empirically distinguishable from classical test theory, item response theory and Bayesian IRT.
The second claim, developed in Section 5, holds that civic identity constitutes a measurable domain of the social self-model, parameterised by five dimensions organised in a three-tier hierarchy. Sovereignty Orientation (the apex parameter encoding the self-model’s predictions about its own Markov blanket) conditions the meaning and behavioural consequences of all lower parameters. The novelty resides not in the discovery of new personality traits but in the derivation of civic parameters from the formal requirements of a social self-model under the FEP.
The third claim, developed in Section 6.2, holds that belonging is formally equivalent to low social free energy, defined as the divergence between an individual’s civic self-model predictions and a community’s observable governance behaviour. This formulation provides a computational neuroscientific basis for the phenomenological and sociological concept of belonging, explaining why person-environment alignment predicts wellbeing (Kristof-Brown, Zimmerman & Johnson, 2005; Fulmer et al., 2010) through the metabolic and cognitive costs of sustained prediction error.
The fourth claim, developed in Section 6.1, holds that formal governance structures (constituted polities with enforceable boundaries, internal governance mechanisms and external interfaces) create genuine Markov blankets at the social level. This renders multi-scale active inference non-metaphorical for such systems whilst conceding that informal social groups lack the structural properties required for genuine blanket formation, constituting the paper’s answer to the multi-scale FEP tautology critique advanced by Colombo & Wright (2021).
The fifth claim, developed alongside the third in Section 6.2, holds that the TO/FOR/BY taxonomy provides a formal distinction between governance modes grounded in which term of the social free energy equation is minimised. TO-mode governance reduces social free energy through evidence restriction. FOR-mode governance reduces it through external active inference. BY-mode governance reduces it through genuine perceptual inference and self-directed active inference. This taxonomy provides formal conditions under which the conformity objection can be empirically investigated, though it does not resolve the first-person epistemic problem for agents embedded in TO-mode environments. The taxonomy also provides the paper’s answer to the question of who governs the measurement. Governance-constituted Markov blankets determine whether the measurement serves the measured individual or the measuring authority. This answer rests on normative premises, specifically the premise that BY-mode governance is preferable to TO-mode governance, a premise grounded in the values of autonomy and epistemic integrity rather than derived from the free energy equations themselves. The paper defends this normative commitment as well-grounded in liberal political philosophy without claiming it is a formal consequence of the mathematics.
These five claims are offered together because they constitute a unified argument, not because each is equally well-supported. The first claim rests on the most established theoretical and empirical foundations. The second and third extend predictive processing into territory where the framework’s computational precision confronts the irreducible complexity of social life. The fourth and fifth are the most speculative, applying the FEP at a scale where the boundary between genuine mechanism and useful metaphor is contested. The paper has been candid about these gradations of confidence, and the empirical programme is designed to test the claims in order of ascending theoretical risk.
A further consequence of the identity-as-inference framework warrants note. If the functional self-model is maintained through precision-weighted prediction error minimisation, then compromised or manipulated self-models should produce detectable statistical anomalies, disrupting the precision structure, coherence patterns and calibration dynamics that characterise an intact functional self-model. This prediction has been confirmed empirically for synthetic systems, where the four-signature protocol developed in Morrison (2026f) achieves perfect discrimination (area under the receiver operating characteristic curve of 1.0) between baseline and adversarially compromised conditions across three distinct attack modalities in the two models first tested, and mean areas of 0.956 (injection) and 0.918 (hijacking) across 37 aligned models, with gradual identity drift detectable from the subtle modification stage (Morrison, 2026g). The practical implication is that the inferential framework developed here offers not merely a measurement approach but a security instrument, capable of detecting identity compromise through disruption of the statistical regularities that an intact self-model maintains.
Six predictions have been specified, each carrying falsification conditions. If metacognitive accuracy does not moderate psychometric reliability (Prediction 1), the constitutive role of reflection in measurement is refuted. If item-specific multi-channel physiological signatures are not detectable or not predictable from independent precision assessment (Prediction 2), the items-as-PE-generators account lacks empirical support. If non-Gaussian distributional features do not predict outcomes beyond point estimates (Prediction 3), the distributional measurement claim is empirically vacuous. If civic parameter alignment does not predict belonging outcomes beyond existing person-environment fit measures (Prediction 4), the civic identity dimensions have not earned their theoretical overhead. If Sovereignty Orientation does not moderate the behavioural consequences of lower-level civic parameters (Prediction 5), the hierarchical structure is falsified. If TO-mode and BY-mode communities do not show the predicted divergence in information-theoretic structure when information restrictions are lifted (Prediction 6), the TO/FOR/BY taxonomy lacks empirical grounding. These are genuine risks, and the paper does not know whether any of them will survive empirical test.
The paper’s broadest claim, offered with appropriate tentativeness, is that identity infrastructure constitutes applied consciousness science. The technologies of identity measurement, matching and governance that are emerging in the computational age are, whether their architects recognise it or not, technologies that operationalise specific theories of what a self is, how it should be known and by whom. Making those theories explicit, subjecting them to the standards of precision, falsifiability and ethical scrutiny that consciousness science at its best demands, is an urgent practical necessity. Predictive processing offers a framework within which the relevant theories can be stated precisely, tested rigorously and governed transparently through the TO/FOR/BY taxonomy and governance-constituted Markov blankets. Whether the specific claims of this paper survive scrutiny matters less than whether the programme of rendering identity infrastructure scientifically and ethically accountable succeeds.
Positionality Statement
The theoretical framework developed in this paper was not produced in purely theoretical circumstances, and intellectual honesty requires a disclosure more forthcoming than academic convention typically demands. The paper’s author is the founder and engineering lead of ~Alter, an identity infrastructure protocol whose engine instantiates the framework this paper proposes, inferring identity from manifestation in observed use and retaining a psychometric assessment as a calibration prior. The formal account of measurement in Sections 2 and 4 holds under continuous ambient observation as readily as under a discrete assessment. Workforce matching is among the first verticals built on the protocol. The protocol’s revenue model depends on the validity of the measurement approach this paper proposes. The civic identity dimensions described in Section 5.3 are specified within the framework and are not deployed as a live civic read, as that deployment is held pending resolution of the social-scoring hazard raised in Section 6.5. The cooperative governance architecture described in Morrison (2026a, Section 6) is the author’s own business structure. The TO/FOR/BY taxonomy formalised in Section 6.2 is a governance-mode analytic that bears on the protocol’s consent architecture, which is the graduated, revocable consent stack together with the constraint clauses governing inference. The relationship between this paper and its author’s commercial enterprise is not incidental.
The intellectual history is candidly bidirectional. Certain theoretical claims were formulated first and subsequently implemented, whilst others emerged from the practical engineering constraints of building a measurement system and were formalised only after the engineering work revealed their necessity. The covariance update equation of Section 2.5, for instance, was motivated by the observation that protocol users’ response reliability varied with their experience of the assessment, an observation that required a formal account of precision accumulation. That assessment has since become a calibration prior, and the history of the formalism is unaffected. The TO/FOR/BY taxonomy was motivated by the need to distinguish the protocol’s consent architecture from the extractive data practices of existing identity measurement systems, a distinction that demanded a formal grounding in the free energy framework rather than a merely normative assertion. Engineering constraints influenced some theoretical choices, and the influence was bidirectional in a manner that the author cannot fully disentangle. The theoretical claims developed in dialogue with practical engineering constraints, and the relationship between theory and application is irreducibly intertwined.
This situation creates obvious risks of confirmation bias, motivated reasoning and the tailoring of theoretical claims to serve commercial interests. The reader is entitled to ask whether the theoretical framework serves the commercial enterprise or the reverse, and the empirical programme of Section 8 is designed to make this question answerable. If the predictions fail, the framework fails regardless of its commercial implementation. The paper has attempted to mitigate the conflict-of-interest risks through several structural commitments. The falsification conditions of Section 8 are specified with sufficient precision that independent researchers can test all six predictions without the author’s involvement. The recommendation for independent validation applies to all six predictions, with particular emphasis on Predictions 4 and 5, which are the predictions most directly tied to the protocol’s commercial value proposition. The objections engaged in Section 7 include the strongest versions the author can construct, including several (the power critique of Section 7.6, the eugenics engagement of Section 7.7, the employment harms of Section 7.8) that identify limitations the framework’s safeguards cannot fully resolve. The paper’s claims are offered as hypotheses to be tested, not as established facts to be implemented.
The author holds no current institutional affiliation, having departed the academic system before completing a doctoral programme. This places the paper outside the legitimation structures of credentialled expertise, and the reader is entitled to weigh this when evaluating its claims. The paper’s response is that the claims stand or fall on their evidential merits, the precision of their formalism and the specificity of their predictions, not on the credentials of their author.
References
Allen, T. A., & DeYoung, C. G. (2017). Personality neuroscience and the Five Factor Model. In T. A. Widiger (Ed.), The Oxford handbook of the Five Factor Model (pp. 319–352). Oxford University Press.
Altemeyer, B. (1996). The authoritarian specter. Harvard University Press.
Andrews, M. (2021). The math is not the territory: Navigating the free energy principle. Biology & Philosophy, 36(4), 30. https://doi.org/10.1007/s10539‑021‑09807‑0
Apps, M. A. J., & Tsakiris, M. (2014). The free-energy self: A predictive coding account of self-recognition. Neuroscience & Biobehavioral Reviews, 41, 85–97. https://doi.org/10.1016/j.neubiorev.2013.01.029
Arendt, H. (1958). The human condition. University of Chicago Press.
Asparouhov, T., Hamaker, E. L., & Muthén, B. (2018). Dynamic structural equation models. Structural Equation Modeling, 25(3), 359–388. https://doi.org/10.1080/10705511.2017.1406803
Baars, B. J. (1988). A cognitive theory of consciousness. Cambridge University Press.
Bandura, A. (1997). Self-efficacy: The exercise of control. W. H. Freeman.
Bateman, A., & Fonagy, P. (2016). Mentalization-based treatment for personality disorders: A practical guide. Oxford University Press. https://doi.org/10.1093/med:psych/9780199680375.001.0001
Berzonsky, M. D., & Sullivan, C. (1992). Social-cognitive aspects of identity style: Need for cognition, experiential openness, and introspection. Journal of Adolescent Research, 7(2), 140–155. https://doi.org/10.1177/074355489272002
Bishop, B. (2008). The big sort: Why the clustering of like-minded America is tearing us apart. Houghton Mifflin.
Boehm, C. (1999). Hierarchy in the forest: The evolution of egalitarian behavior. Harvard University Press.
Boehm, C. (2012). Moral origins: The evolution of virtue, altruism, and shame. Basic Books.
Borsboom, D., & Cramer, A. O. J. (2013). Network analysis: An integrative approach to the structure of psychopathology. Annual Review of Clinical Psychology, 9, 91–121. https://doi.org/10.1146/annurev‑clinpsy‑050212‑185608
Borsboom, D., Mellenbergh, G. J., & van Heerden, J. (2003). The theoretical status of latent variables. Psychological Review, 110(2), 203–219. https://doi.org/10.1037/0033‑295X.110.2.203
Borsboom, D., Mellenbergh, G. J., & van Heerden, J. (2004). The concept of validity. Psychological Review, 111(4), 1061–1071. https://doi.org/10.1037/0033‑295X.111.4.1061
Bouizegarene, N., Ramstead, M. J. D., Constant, A., Friston, K. J., & Kirmayer, L. J. (2024). Narrative as active inference: An integrative account of cognitive and social functions in adaptation. Frontiers in Psychology, 15, 1345480. https://doi.org/10.3389/fpsyg.2024.1345480
Brown, R., Lau, H., & LeDoux, J. E. (2019). Understanding the higher-order approach to consciousness. Trends in Cognitive Sciences, 23(9), 754–768. https://doi.org/10.1016/j.tics.2019.06.009
Bruineberg, J., Kiverstein, J., & Rietveld, E. (2018). The anticipating brain is not a scientist: The free-energy principle from an ecological-enactive perspective. Synthese, 195(6), 2417–2444. https://doi.org/10.1007/s11229‑016‑1239‑1
Butler, J. (1990). Gender trouble: Feminism and the subversion of identity. Routledge.
Butler, J. (1993). Bodies that matter: On the discursive limits of ‘sex.’ Routledge.
Campbell, J. D., Trapnell, P. D., Heine, S. J., Katz, I. M., Lavallee, L. F., & Lehman, D. R. (1996). Self-concept clarity: Measurement, personality correlates, and cultural boundaries. Journal of Personality and Social Psychology, 70(1), 141–156. https://doi.org/10.1037/0022‑3514.70.1.141
Carens, J. H. (2013). The ethics of immigration. Oxford University Press.
Colombo, M., & Wright, C. (2021). First principles in the life sciences: The free-energy principle, organicism, and mechanism. Synthese, 198(Suppl 14), 3463–3488. https://doi.org/10.1007/s11229‑018‑01932‑w
Costa, P. T., Jr., & McCrae, R. R. (1992). Revised NEO Personality Inventory (NEO-PI-R) and NEO Five-Factor Inventory (NEO-FFI) professional manual. Psychological Assessment Resources.
Cramer, A. O. J., van Borkulo, C. D., Giltay, E. J., van der Maas, H. L. J., Kendler, K. S., Scheffer, M., & Borsboom, D. (2016). Major depression as a complex dynamic system. PLOS ONE, 11(12), e0167490. https://doi.org/10.1371/journal.pone.0167490
De Boeck, P., Cho, S.-J., & Wilson, M. (2011). Explanatory item response models. In S. E. Embretson (Ed.), Measuring psychological constructs (pp. 43–69). American Psychological Association.
de Montjoye, Y.-A., Hidalgo, C. A., Verleysen, M., & Blondel, V. D. (2013). Unique in the crowd: The privacy bounds of human mobility. Scientific Reports, 3, 1376. https://doi.org/10.1038/srep01376
de Waal, F. B. M. (1996). Good natured: The origins of right and wrong in humans and other animals. Harvard University Press.
Deci, E. L., & Ryan, R. M. (2000). The ‘what’ and ‘why’ of goal pursuits: Human needs and the self-determination of behavior. Psychological Inquiry, 11(4), 227–268. https://doi.org/10.1207/S15327965PLI1104_01
Dehaene, S., & Naccache, L. (2001). Towards a cognitive neuroscience of consciousness: Basic evidence and a workspace framework. Cognition, 79(1–2), 1–37. https://doi.org/10.1016/S0010‑0277(00)00123‑2
DeYoung, C. G., Quilty, L. C., & Peterson, J. B. (2007). Between facets and domains: 10 aspects of the Big Five. Journal of Personality and Social Psychology, 93(5), 880–896. https://doi.org/10.1037/0022‑3514.93.5.880
Embretson, S. E. (1983). Construct validity: Construct representation versus nomothetic span. Psychological Bulletin, 93(1), 179–197. https://doi.org/10.1037/0033‑2909.93.1.179
Embretson, S. E. (1998). A cognitive design system approach to generating valid tests: Application to abstract reasoning. Psychological Methods, 3(3), 380–396. https://doi.org/10.1037/1082‑989X.3.3.380
Embretson, S. E., & Reise, S. P. (2000). Item response theory for psychologists. Lawrence Erlbaum Associates.
Erikson, E. H. (1968). Identity: Youth and crisis. W. W. Norton.
Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive-developmental inquiry. American Psychologist, 34(10), 906–911. https://doi.org/10.1037/0003‑066X.34.10.906
Fleeson, W. (2001). Toward a structure- and process-integrated view of personality: Traits as density distributions of states. Journal of Personality and Social Psychology, 80(6), 1011–1027. https://doi.org/10.1037/0022‑3514.80.6.1011
Fleeson, W., & Jayawickreme, E. (2015). Whole Trait Theory. Journal of Research in Personality, 56, 82–92. https://doi.org/10.1016/j.jrp.2014.10.009
Fleming, S. M., & Dolan, R. J. (2012). The neural basis of metacognitive ability. Philosophical Transactions of the Royal Society B, 367(1594), 1338–1349. https://doi.org/10.1098/rstb.2011.0417
Fleming, S. M., & Lau, H. C. (2014). How to measure metacognition. Frontiers in Human Neuroscience, 8, Article 443. https://doi.org/10.3389/fnhum.2014.00443
Fleming, S. M., Ryu, J., Golfinos, J. G., & Blackmon, K. E. (2014). Domain-specific impairment in metacognitive accuracy following anterior prefrontal lesions. Brain, 137(10), 2811–2822. https://doi.org/10.1093/brain/awu221
Fleming, S. M., Weil, R. S., Nagy, Z., Dolan, R. J., & Rees, G. (2010). Relating introspective accuracy to individual differences in brain structure. Science, 329(5998), 1541–1543. https://doi.org/10.1126/science.1191883
Fonagy, P., Gergely, G., Jurist, E. L., & Target, M. (2002). Affect regulation, mentalization, and the development of the self. Other Press.
Foucault, M. (1977). Discipline and punish: The birth of the prison (A. Sheridan, Trans.). Vintage Books.
Fox, C. A., McDonogh, A., Donegan, K. R., Teckentrup, V., Crossen, R. J., Hanlon, A. K., Gallagher, E., Rouault, M., & Gillan, C. M. (2024). Reliable, rapid, and remote measurement of metacognitive bias. Scientific Reports, 14, 14941. https://doi.org/10.1038/s41598‑024‑64900‑0
Fox, J.-P. (2010). Bayesian item response modeling: Theory and applications. Springer. https://doi.org/10.1007/978‑1‑4419‑0742‑4
Friston, K. J. (2008). Hierarchical models in the brain. PLoS Computational Biology, 4(11), e1000211. https://doi.org/10.1371/journal.pcbi.1000211
Friston, K. J. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138. https://doi.org/10.1038/nrn2787
Fulmer, C. A., Gelfand, M. J., Kruglanski, A. W., Kim-Prieto, C., Diener, E., Pierro, A., & Higgins, E. T. (2010). On ‘feeling right’ in cultural contexts: How person-culture match affects self-esteem and subjective well-being. Psychological Science, 21(11), 1563–1569. https://doi.org/10.1177/0956797610384742
Gallese, V. (2001). The ‘shared manifold’ hypothesis: From mirror neurons to empathy. Journal of Consciousness Studies, 8(5–7), 33–50.
Gallese, V. (2005). Embodied simulation: From neurons to phenomenal experience. Phenomenology and the Cognitive Sciences, 4(1), 23–48. https://doi.org/10.1007/s11097‑005‑4737‑z
Gallese, V., & Sinigaglia, C. (2011). What is so special about embodied simulation? Trends in Cognitive Sciences, 15(11), 512–519. https://doi.org/10.1016/j.tics.2011.09.003
Hacking, I. (1986). Making up people. In T. C. Heller, M. Sosna, & D. E. Wellbery (Eds.), Reconstructing individualism (pp. 222–236). Stanford University Press.
Haidt, J. (2012). The righteous mind: Why good people are divided by politics and religion. Pantheon Books.
Hamaker, E. L., Asparouhov, T., Brose, A., Schmiedek, F., & Muthén, B. (2018). At the frontiers of modeling intensive longitudinal data: Dynamic structural equation models for the affective measurements from the COGITO study. Multivariate Behavioral Research, 53(6), 820–841. https://doi.org/10.1080/00273171.2018.1446819
Han, S., Northoff, G., Vogeley, K., Wexler, B. E., Kitayama, S., & Varnum, M. E. W. (2013). A cultural neuroscience approach to the biosocial nature of the human brain. Annual Review of Psychology, 64, 335–359. https://doi.org/10.1146/annurev‑psych‑071112‑054629
Henrich, J. (2020). The WEIRDest people in the world: How the West became psychologically peculiar and particularly prosperous. Farrar, Straus and Giroux.
Hohwy, J. (2013). The predictive mind. Oxford University Press.
Hohwy, J. (2025). The self-evidencing agent. MIT Press.
Hutto, D. D., & Myin, E. (2013). Radicalizing enactivism: Basic minds without content. MIT Press.
Hutto, D. D., & Myin, E. (2017). Evolving enactivism: Basic minds meet content. MIT Press.
Inhelder, B., & Piaget, J. (1958). The growth of logical thinking from childhood to adolescence. Basic Books.
Isomura, T., Kotani, K., Jimbo, Y., & Friston, K. J. (2023). Experimental validation of the free-energy principle with in vitro neural networks. Nature Communications, 14, 4547. https://doi.org/10.1038/s41467‑023‑40141‑z
Jost, J. T., Federico, C. M., & Napier, J. L. (2009). Political ideology: Its structure, functions, and elective affinities. Annual Review of Psychology, 60, 307–337. https://doi.org/10.1146/annurev.psych.60.110707.163600
Kaaronen, R. O. (2018). A theory of predictive dissonance: Predictive processing presents a new take on cognitive dissonance. Frontiers in Psychology, 9, 2218. https://doi.org/10.3389/fpsyg.2018.02218
Kaplan, J. T., Gimbel, S. I., & Harris, S. (2016). Neural correlates of maintaining one’s political beliefs in the face of counterevidence. Scientific Reports, 6, 39589. https://doi.org/10.1038/srep39589
Karabatsos, G. (2017). A menu-driven software package of Bayesian nonparametric (and parametric) mixed models for regression analysis and density estimation. Behavior Research Methods, 49(1), 335–362. https://doi.org/10.3758/s13428‑016‑0711‑7
Kiebel, S. J., Daunizeau, J., & Friston, K. J. (2008). A hierarchy of time-scales and the brain. PLoS Computational Biology, 4(11), e1000209. https://doi.org/10.1371/journal.pcbi.1000209
Kirchhoff, M. D., Parr, T., Palacios, E., Friston, K. J., & Kiverstein, J. (2018). The Markov blankets of life: Autonomy, active inference and the free energy principle. Journal of the Royal Society Interface, 15(138), Article 20170792. https://doi.org/10.1098/rsif.2017.0792
Kristof-Brown, A. L., Zimmerman, R. D., & Johnson, E. C. (2005). Consequences of individuals’ fit at work: A meta-analysis of person-job, person-organization, person-group, and person-supervisor fit. Personnel Psychology, 58(2), 281–342. https://doi.org/10.1111/j.1744‑6570.2005.00672.x
Lee, K., & Ashton, M. C. (2004). Psychometric properties of the HEXACO personality inventory. Multivariate Behavioral Research, 39(2), 329–358. https://doi.org/10.1207/s15327906mbr3902_8
Levy, R., & Mislevy, R. J. (2016). Bayesian psychometric modeling. CRC Press.
Limanowski, J., & Blankenburg, F. (2013). Minimal self-models and the free energy principle. Frontiers in Human Neuroscience, 7, Article 547. https://doi.org/10.3389/fnhum.2013.00547
Lord, F. M., & Novick, M. R. (1968). Statistical theories of mental test scores. Addison-Wesley.
Lysaker, P. H., Gagen, E., Klion, R., Zalzala, A., Vohs, J., Faith, L. A., Leonhardt, B., Hamm, J., & Hasson-Ohayon, I. (2020). Metacognitive reflection and insight therapy: A recovery-oriented treatment approach for psychosis. Psychology Research and Behavior Management, 13, 331–341. https://doi.org/10.2147/PRBM.S198628
Maniscalco, B., & Lau, H. (2012). A signal detection theoretic approach for estimating metacognitive sensitivity from confidence ratings. Consciousness and Cognition, 21(1), 422–430. https://doi.org/10.1016/j.concog.2011.09.021
Marcia, J. E. (1966). Development and validation of ego-identity status. Journal of Personality and Social Psychology, 3(5), 551–558. https://doi.org/10.1037/h0023281
Markon, K. E., Krueger, R. F., & Watson, D. (2005). Delineating the structure of normal and abnormal personality: An integrative hierarchical approach. Journal of Personality and Social Psychology, 88(1), 139–157. https://doi.org/10.1037/0022‑3514.88.1.139
Mathys, C., Daunizeau, J., Friston, K. J., & Stephan, K. E. (2011). A Bayesian foundation for individual learning under uncertainty. Frontiers in Human Neuroscience, 5, 39. https://doi.org/10.3389/fnhum.2011.00039
Mathys, C. D., Lomakina, E. I., Daunizeau, J., Iglesias, S., Brodersen, K. H., Friston, K. J., & Stephan, K. E. (2014). Uncertainty in perception and the Hierarchical Gaussian Filter. Frontiers in Human Neuroscience, 8, 825. https://doi.org/10.3389/fnhum.2014.00825
McAdams, D. P. (2001). The psychology of life stories. Review of General Psychology, 5(2), 100–122. https://doi.org/10.1037/1089‑2680.5.2.100
Meijer, R. R., & Sijtsma, K. (2001). Methodology review: Evaluating person fit. Applied Psychological Measurement, 25(2), 107–135. https://doi.org/10.1177/01466210122031957
Menon, V. (2023). 20 years of the default mode network: A review and synthesis. Neuron, 111(16), 2469–2487. https://doi.org/10.1016/j.neuron.2023.04.023
Metcalfe, J. (2009). Metacognitive judgments and control of study. Current Directions in Psychological Science, 18(3), 159–163. https://doi.org/10.1111/j.1467‑8721.2009.01628.x
Metzinger, T. (2003). Being no one: The self-model theory of subjectivity. MIT Press.
Metzinger, T. (2004). Being no one: The self-model theory of subjectivity (Paperback ed.). MIT Press.
Michell, J. (2000). Normal science, pathological science and psychometrics. Theory & Psychology, 10(5), 639–667. https://doi.org/10.1177/0959354300105004
Michell, J. (2008). Is psychometrics pathological science? Measurement: Interdisciplinary Research and Perspectives, 6(1–2), 7–24. https://doi.org/10.1080/15366360802035489
Miyazaki, K., & Hoshino, T. (2009). A Bayesian semiparametric item response model with Dirichlet process priors. Psychometrika, 74(3), 375–393. https://doi.org/10.1007/s11336‑008‑9108‑6
Morgeson, F. P., Campion, M. A., Dipboye, R. L., Hollenbeck, J. R., Murphy, K., & Schmitt, N. (2007). Reconsidering the use of personality tests in personnel selection contexts. Personnel Psychology, 60(3), 683–729. https://doi.org/10.1111/j.1744‑6570.2007.00089.x
Morrison, B. (2026a). Jus identitatis: Toward post-geographic sovereignty through measured identity alignment [Preprint]. Figshare. https://doi.org/10.6084/m9.figshare.31794784
Morrison, B. (2026d). Generative psychometrics: Measurement theory for self-reflective constructs [Preprint]. Figshare. https://doi.org/10.6084/m9.figshare.31812982
Morrison, B. (2026f). The self-model test: A measurement protocol for synthetic self-models [Preprint]. Figshare. https://doi.org/10.6084/m9.figshare.31813006
Morrison, B. (2026g). Empirical validation of the identity-as-inference framework [Preprint]. Figshare. https://doi.org/10.6084/m9.figshare.31951644
Motyl, M., Iyer, R., Oishi, S., Trawalter, S., & Nosek, B. A. (2014). How ideological migration geographically segregates groups. Journal of Experimental Social Psychology, 51, 1–14. https://doi.org/10.1016/j.jesp.2013.10.010
Moutoussis, M., Fearon, P., El-Deredy, W., Dolan, R. J., & Friston, K. J. (2014). Bayesian inferences about the self (and others): A review. Consciousness and Cognition, 25, 67–76. https://doi.org/10.1016/j.concog.2014.01.009
Murphy, R. O., Ackermann, K. A., & Handgraaf, M. J. J. (2011). Measuring social value orientation. Judgment and Decision Making, 6(8), 771–781.
Nordahl, H., Hjemdal, O., Hagen, R., Nordahl, H. M., & Wells, A. (2019). What lies beneath trait-anxiety? Testing the self-regulatory executive function model of vulnerability. Frontiers in Psychology, 10, 122. https://doi.org/10.3389/fpsyg.2019.00122
Ostrom, E. (1990). Governing the commons: The evolution of institutions for collective action. Cambridge University Press.
Page, S. E. (2007). The difference: How the power of diversity creates better groups, firms, schools, and societies. Princeton University Press.
Peirce, C. S. (1878). How to make our ideas clear. Popular Science Monthly, 12, 286–302.
Phillips, K. W. (2014). How diversity works. Scientific American, 311(4), 42–47.
Popper, K. R. (1959). The logic of scientific discovery. Hutchinson.
Putnam, R. D. (2000). Bowling alone: The collapse and revival of American community. Simon & Schuster.
Raghavan, M., Barocas, S., Kleinberg, J., & Levy, K. (2020). Mitigating bias in algorithmic hiring: Evaluating claims and practices. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency (pp. 469–481). Association for Computing Machinery. https://doi.org/10.1145/3351095.3372828
Ramstead, M. J. D., Badcock, P. B., & Friston, K. J. (2018). Answering Schrödinger’s question: A free-energy formulation. Physics of Life Reviews, 24, 1–16. https://doi.org/10.1016/j.plrev.2017.09.001
Ramstead, M. J. D., Veissière, S. P. L., & Kirmayer, L. J. (2016). Cultural affordances: Scaffolding local worlds through shared intentionality and regimes of attention. Frontiers in Psychology, 7, 1090. https://doi.org/10.3389/fpsyg.2016.01090
Rawls, J. (1971). A theory of justice. Harvard University Press.
Rivera, L. A. (2012). Hiring as cultural matching: The case of elite professional service firms. American Sociological Review, 77(6), 999–1022. https://doi.org/10.1177/0003122412463213
Rivera, L. A. (2015). Pedigree: How elite students get elite jobs. Princeton University Press.
Roberts, B. W., & DelVecchio, W. F. (2000). The rank-order consistency of personality traits from childhood to old age: A quantitative review of longitudinal studies. Psychological Bulletin, 126(1), 3–25. https://doi.org/10.1037/0033‑2909.126.1.3
Roberts, D. (2012). Fatal invention: How science, politics, and big business re-create race in the twenty-first century. The New Press.
Rosenthal, D. M. (1986). Two concepts of consciousness. Philosophical Studies, 49(3), 329–359.
Rosenthal, D. M. (2005). Consciousness and mind. Oxford University Press.
Sapolsky, R. M. (2005). The influence of social hierarchy on primate health. Science, 308(5722), 648–652. https://doi.org/10.1126/science.1106477
Schwartz, S. H. (2012). An overview of the Schwartz theory of basic values. Online Readings in Psychology and Culture, 2(1), 11. https://doi.org/10.9707/2307‑0919.1116
Seth, A. K. (2013). Interoceptive inference, emotion, and the embodied self. Trends in Cognitive Sciences, 17(11), 565–573. https://doi.org/10.1016/j.tics.2013.09.007
Seth, A. K., & Friston, K. J. (2016). Active interoceptive inference and the emotional brain. Philosophical Transactions of the Royal Society B, 371(1708), 20160007. https://doi.org/10.1098/rstb.2016.0007
Seth, A. K., & Tsakiris, M. (2018). Being a beast machine: The somatic basis of selfhood. Trends in Cognitive Sciences, 22(11), 969–981. https://doi.org/10.1016/j.tics.2018.08.008
Shachar, A. (2009). The birthright lottery: Citizenship and global inequality. Harvard University Press.
Sidanius, J., & Pratto, F. (1999). Social dominance: An intergroup theory of social hierarchy and oppression. Cambridge University Press.
Singelis, T. M. (1994). The measurement of independent and interdependent self-construals. Personality and Social Psychology Bulletin, 20(5), 580–591. https://doi.org/10.1177/0146167294205014
Stuss, D. T., & Alexander, M. P. (2000). Executive functions and the frontal lobes: A conceptual view. Psychological Research, 63(3–4), 289–298. https://doi.org/10.1007/s004269900007
Taylor, C. (1989). Sources of the self: The making of the modern identity. Harvard University Press.
Tetlock, P. E. (1983). Cognitive style and political ideology. Journal of Personality and Social Psychology, 45(1), 118–126. https://doi.org/10.1037/0022‑3514.45.1.118
Thom, R. (1975). Structural stability and morphogenesis. Benjamin.
Thompson, E. (2007). Mind in life: Biology, phenomenology, and the sciences of mind. Harvard University Press.
Tomasello, M. (2014). A natural history of human thinking. Harvard University Press.
Tomasello, M. (2019). Becoming human: A theory of ontogeny. Harvard University Press.
Tourangeau, R., Rips, L. J., & Rasinski, K. (2000). The psychology of survey response. Cambridge University Press.
Tsakiris, M. (2017). The multisensory basis of the self: From body to identity to others. Quarterly Journal of Experimental Psychology, 70(4), 597–609. https://doi.org/10.1080/17470218.2016.1181768
Tuomela, R. (2013). Social ontology: Collective intentionality and group agents. Oxford University Press.
van der Maas, H. L. J., & Molenaar, P. C. M. (1992). Stagewise cognitive development: An application of catastrophe theory. Psychological Review, 99(3), 395–417.
Veissière, S. P. L., Constant, A., Ramstead, M. J. D., Friston, K. J., & Kirmayer, L. J. (2020). Thinking through other minds: A variational approach to cognition and culture. Behavioral and Brain Sciences, 43, Article e90. https://doi.org/10.1017/S0140525X19001213
Webster, D. M., & Kruglanski, A. W. (1994). Individual differences in need for cognitive closure. Journal of Personality and Social Psychology, 67(6), 1049–1062. https://doi.org/10.1037/0022‑3514.67.6.1049
Welzel, C. (2013). Freedom rising: Human empowerment and the quest for emancipation. Cambridge University Press.
Wetzel, E., Lüdtke, O., Zettler, I., & Böhnke, J. R. (2016). The stability of extreme response style and acquiescence over 8 years. Assessment, 23(3), 279–291. https://doi.org/10.1177/1073191115583714
Wheeler, N. E., Allidina, S., Long, E. U., Schneider, S. P., Haas, I. J., & Cunningham, W. A. (2020). Ideology and predictive processing: Coordination, bias, and polarization in socially constrained error minimization. Current Opinion in Behavioral Sciences, 34, 192–198. https://doi.org/10.1016/j.cobeha.2020.05.002
Ylvisaker, M., & Feeney, T. (2000). Reconstruction of identity after traumatic brain injury. Brain Impairment, 1(1), 12–28.
Young, I. M. (1990). Justice and the politics of difference. Princeton University Press.
Zahavi, D. (2005). Subjectivity and selfhood: Investigating the first-person perspective. MIT Press.
Zahavi, D. (2014). Self and other: Exploring subjectivity, empathy, and shame. Oxford University Press.