Interoceptive control
The wiki has been circling this for several ingests and can now state it directly:
Almost every interoceptive construct in the field is defined by perception, and almost every theory of interoception is a theory of control.
Four taxonomies are catalogued on interoceptive-taxonomy — Farb’s seven, Garfinkel’s three, Khalsa’s eight, Berntson’s seven. Between them they partition detection, attention, magnitude, discrimination, accuracy, sensitivity, sensibility, insight, coherence, awareness. Every one of those is something the brain reads. Exactly one construct across all four lists is efferent — Farb et al.’s regulation — and until Allen (2026) there was no task for it.
Meanwhile the theoretical side has been about control the whole time. active-inference resolves prediction error by changing the body, not just the model. allostasis is anticipatory regulation. Chen et al. proposed a seventh definition of interoception that folds descending regulation inside the term, complete with “interoceptive effectors.” Berntson & Khalsa argue the unit of analysis is a circuit, not a channel, precisely because afference cannot be described apart from the efferent systems that change what is sensed. Petzschner et al. put action selection on the middle arc of the three-arc loop as a peer of state inference.
So the mismatch is not subtle. The field’s models are closed-loop; its measurements are open-loop and afferent-only.
Why the gap persisted: you cannot score what cannot be driven
The reason is not oversight. It is that most interoceptive channels admit no voluntary control at all, so there is nothing to score.
You cannot decide to change your gastric motility, your bladder filling rate, your immune signalling, your blood glucose, or (much) your heart rate. A control task requires a channel the person can drive on demand, and there is essentially one: respiration.
This produces an awkward position the wiki should hold explicitly rather than resolve:
- Respiration is the least representative interoceptive channel, precisely because it is voluntarily drivable (respiratory-interoception, and Weng et al.’s own concession that “not all systems can be as readily consciously manipulated”).
- Respiration is the only channel on which interoceptive control is measurable at all.
Both are true, and together they imply something uncomfortable: interoceptive control may not be a general construct. It may be a fact about breathing. Anything measured with the respiratory-tracking-task generalizes to the gut or the bladder only if the voluntary-drivability property was doing no work — and in this case that property is the task.
The wiki’s position: treat interoceptive control as demonstrated for respiration and unestablished as a person-level trait, pending exactly the evidence interoceptive-taxonomy already flags as missing for accuracy.
Update, and it is not the resolution this paragraph expected. Banellis et al. (2026) — named here as the pending evidence — has now been read, and cardiac and respiratory perception do decorrelate: sensitivity, precision and metacognitive efficiency all uncorrelated at N = 241. That makes a general control construct less likely by analogy and tests nothing about control directly, because there is no control task in any channel but this one. The analogy is suggestive rather than probative: perception and control could perfectly well have different generality profiles, and nobody knows, because measuring control in a second channel would require a second drivable channel. So this section’s question is not merely open but currently unaskable with existing instruments — a stronger statement than it could make before. See is-interoception-domain-general.
Two things that are not this
Kept separate deliberately, because merging them dissolves the concept:
Not allostasis. Allostatic regulation is descending, automatic, and unavailable to report or instruction. Nobody scores a participant on their baroreflex. Interoceptive control as used here is deliberate — instructed, effortful, and improvable within a session.
Not slow-breathing. Device-guided slow breathing changes an interoceptive channel’s state but is scored on downstream physiology (MSNA, blood pressure, baroreflex sensitivity), not on how well the person did it. The wiki already records it as “a manipulation of an interoceptive pathway rather than of interoception,” and it needs no awareness at all. The control construct requires that the person’s performance be the measured quantity. Notably, the tracking task supplies exactly the missing fidelity measure for slow breathing.
The dissociation on offer
The substantive proposal from Allen is that control decomposes into two things:
- Control ability — how accurately the target waveform can be produced under veridical feedback.
- Sensorimotor flexibility — how much accuracy degrades when the action-to-consequence mapping is distorted, normalized against ability (the perturbation ratio).
These came apart in the validation data: fatigue degraded flexibility while leaving ability intact. If the dissociation survives in real samples, it is a second axis the field has never had, and it maps onto a distinction the predictive-coding pages assume without measuring — executing a policy versus updating the internal model when the policy stops working. Failure to remap is not weakness; it is a stale forward model.
What this would change if it holds
- Contemplative training. does-mindfulness-enhance-interoceptive-accuracy is a repeated null on cardiac accuracy, against practitioners’ insistence that something changed. Control has never been the dependent variable, and breath-focused practice trains it directly by construction. This is the most obvious place the null might be a measurement artefact rather than an absence.
- Panic. cognitive-model-of-panic and anxiety-sensitivity are theories of catastrophic misreading of respiratory sensation. Whether panic also involves a control deficit — worse tracking, or a higher remapping cost when the breath does not respond as expected — is untested and clinically pointed.
- is-more-interoceptive-awareness-better. The debate is stuck partly because “more contact with the body” is measured by attention and sensibility scales that cannot distinguish useful monitoring from anxious monitoring. A control measure sidesteps the question: it scores what the person can do, not how much they attend. Whether that predicts outcomes better than sensibility does is an empirical question the instrument now permits.
- perceptual-inference-as-regulation. That debate asks whether perceptual inference is genuine regulation or collapses into active inference. A task that separates control ability from remapping capacity gives the distinction a behavioural handle it did not have.
The problem at the bottom of it
Stated once more, because everything above depends on it: the one task measuring interoceptive control gives the participant visual feedback about their breathing. Good tracking may reflect good interoception, or good motor control reading a screen. See respiratory-tracking-task for the designs that would separate these — chiefly, removing the feedback.
Until then, “interoceptive control” names a construct whose interoceptive credentials rest on the substrate being controlled rather than on the information doing the controlling.