THEOUTSIDE FACE & INSIDE FACE
A Systematic Mapping of the Return Structure onto Interoceptive Predictive Coding
Predictive coding describes the machinery.
The "return" describes why any of it matters.
One is the outside face. The other is the inside face.
This paper tests whether they are faces of the same process — or whether one is redundant.
The caring gap framework claims that interoceptive predictive coding describes the "outside face" of a process whose "inside face" is the return — a temporal structure of oriented fidelity across interruption that constitutes felt mattering. This complementarity has been stated philosophically in The Arriving Breath, developed clinically in twenty-eight falsifiable predictions, and grounded biologically in The Body Is The First Return.
It has not been tested against the predictive coding literature at the level of technical precision the claim requires. This paper performs that test.
The method: take six core features of Seth's (2012; 2013) interoceptive inference model of conscious presence and map each against its counterpart in the return structure. For each, three questions: Where do they align? Where does the return add something? Where does predictive coding have something the return lacks?
If the mapping produces novel predictions that neither framework generates alone, the complementarity has empirical content. If it does not, the "inside face" is a redescription of what precision-weighting already explains.
| Feature | Alignment | Return adds | PC has |
|---|---|---|---|
| Hierarchical model | Aligned | Temporal hierarchy | Math formalism |
| Both describe nested hierarchy. Seth's is spatial — anatomical. The return's is temporal — organised by interval length: milliseconds at cellular, seconds at heartbeat, minutes at respiration, hours at interpersonal. Disruptions propagate fast→slow, generating Prediction A. The return has no mathematical formalism — its single largest gap. | |||
| Precision-weighting | Aligned | Directionality | Neural locus |
| Precision is a scalar. Fidelity is directional — toward something specific. Saini et al. (2022): DPDR patients showed impaired heartbeat detection with intact confidence. Precision preserved, fidelity collapsed. Generates Prediction B. | |||
| Error suppression | Divergent | Positive presence | HEP data |
| Seth: presence is absence of error. The return: presence is a felt temporal form. Opposite HEP predictions — elevated in DPDR (Seth) vs. reduced (return). A three-group design (DPDR, anxiety, controls) would be decisive. | |||
| Active inference | Complementary | Why it matters | Domain bounds |
| Active inference without the return = thermostat (mechanism without mattering). The return without active inference = ghost (mattering without mechanism). Together: the organism's felt, enacted persistence. | |||
| Anterior insula | Circuit match | Temporal window | Neural data |
| The return's neural counterpart: temporal property of the insula-ACC loop — duration over which past signals integrate. Collapsed window = each heartbeat disconnected from predecessors. Generates Prediction C. | |||
| Emotional inference | Divergent | Grief account | Therapy data |
| Seth: grief = persistent prediction error. The return: grief = the return arriving at absence. Different resolution predictions: model-updating (cognitive restructuring) vs. return-adaptation (repeated encounter with absence in a held context). Effective grief therapy uses the second. | |||
The brain maintains a multi-level generative model that predicts incoming sensory signals. Higher levels generate predictions passed down; lower levels send prediction errors up. Conscious presence arises when this hierarchy successfully suppresses interoceptive prediction errors. The hierarchy is spatial — organised anatomically from brainstem through insula to prefrontal cortex.
The return operates at nested scales: cellular (the sodium-potassium pump), organ-level (the heartbeat), whole-body (body-brain-body loop), interpersonal (the caregiver returning across interruption). Each scale's return is built on and calibrated by the scale below.
Seth's hierarchy is spatial. The return's is temporal — organised by interval length. Disruptions at faster timescales propagate to slower ones before the reverse. The body convergence — thirteen findings from twelve groups in eight countries — is better explained by a temporal hierarchy than a spatial one.
Mathematical specification: prediction error minimisation, Bayesian updating, precision-weighting as inverse variance. The return has no formalism. This is the single largest gap.
The brain assigns precision (confidence) to predictions and signals. DPDR arises when interoceptive precision is aberrantly low. The return's counterpart: fidelity — graded, directional orientation maintained across the interval.
Precision is a scalar. Fidelity is directional — toward something specific. Precision tells you how much confidence; it doesn't tell you what the signal is for. Saini et al. (2022): impaired heartbeat detection with intact confidence. Precision preserved, fidelity collapsed.
Seth: presence = successful suppression of interoceptive prediction errors. Defined negatively — what you get when errors go away. The return: mattering = felt quality of the temporal form itself. Positive characterisation.
If both clinical groups show elevated HEP, precision-weighting is sufficient. If they diverge, the two-failure-modes account is confirmed.
The brain actively changes the body to match predictions. The return's counterpart: the organism enacts its own continuation — maintaining conditions for its own mattering.
Active inference without the return is the thermostat — mechanism without mattering. The return without active inference is a ghost — mattering without mechanism. Together: the organism's felt, enacted persistence. The domain restriction is a strength.
The anterior insular cortex is where predictions meet signals. Jia et al. (2025) confirmed DPDR alters precisely the body-to-brain integration hubs.
The return's neural counterpart: a temporal property of the insula-ACC loop — the duration over which past signals integrate into the current state. Normal window = current heartbeat carries felt trace of previous heartbeats. Collapsed = each heartbeat disconnected from predecessors.
Emotions are configurations of the return's temporal form. Grief is the return arriving at absence. Shame is the return in a membrane that believes it should not exist. Indifference is the return collapsed.
Seth: grief = persistent prediction error. The return: grief = the return arriving at absence. Different resolution: model-updating vs. return-adaptation. Effective grief therapy uses the second, not the first.
Each arises where the frameworks diverge. Each is testable with existing methods and populations.
The Absorption Threat
The nearest competitor is Balar and Kapila's (2025) integrative precision-weighting model — parallel hierarchical systems with arbitration in the anterior insula and ACC. Exactly the circuit proposed as the locus of the return's temporal integration.
If their model derives the same predictions, the complementarity claim weakens. The differentiating tests: Predictions A and C. If symptom onset ordering and temporal integration decay are derivable from precision-weighting alone, the return adds nothing.
If they are not, the temporal dimension is irreducible.
The return adds three things predictive coding does not have: a temporal hierarchy predicting symptom onset ordering, a directional quality separable from precision, and a positive characterisation of presence generating a divergent HEP prediction.
Predictive coding has three things the return needs: mathematical formalism, neural implementation, and domain boundaries.
The mapping demonstrates a method — feature-by-feature alignment between experiential and computational frameworks — offered as a template for any experiential framework claiming to complement predictive coding.
Declarations
No competing interests. No external funding. No original data. Developed in dialogue with AI systems (Claude, Anthropic), disclosed in the interest of transparency.
References
Balar & Kapila (2025) · Craig (2009) · Ferroni et al. (2025) · Friston (2010) · Gatus et al. (2022) · Jia et al. (2025) · Kogura (2026a–d) · Michal et al. (2013) · Pini et al. (2025) · Saini et al. (2022) · Seth (2013) · Seth, Suzuki & Critchley (2012) · Sierra & David (2011) · Zheng et al. (2024a, 2024b)
Full reference list at caring-gap.com
Citation
Kogura, J. S. (2026). The Outside Face and the Inside Face: A Systematic Mapping of the Return Structure onto Interoceptive Predictive Coding. caring-gap.com.