Interoceptive precision

The most theoretically load-bearing quantity in this wiki, and until recently the least measured one.

The theoretical sense

In predictive-coding and interoceptive-inference, a prediction error’s influence on belief is scaled by its precision — the confidence the system places in that signal. High precision means the error is taken seriously and the model updates; low precision means it is discounted and the prior stands. This is not a detail of implementation; it is where most of the explanatory work in the framework happens, because the same architecture produces very different behaviour depending on how precision is allocated.

The wiki has been running on it for a long time:

• Selfhood. Seth (2013) reads the finding that lower interoceptive sensitivity predicts greater susceptibility to the rubber-hand-illusion (Tsakiris et al. 2011) as low precision-weighting of interoceptive prediction errors, letting exteroceptive cues dominate the self-model. Quigley et al. later state this as a precision competition between channels, following Ainley et al.’s (2016) “bodily precision.” See experience-of-body-ownership.
• Psychopathology. computational-psychiatry and interoceptive-trait-prediction-error both explain symptoms by mis-set precision — over-weighted priors producing symptoms in the absence of signal, or over-weighted errors producing intolerable bodily noise. Smith et al. (2020) frame transdiagnostic dysfunction as a failure to adapt interoceptive precision.
• Cross-modal difference. Quadt, Critchley & Garfinkel propose that interoceptive modalities are precision-weighted differently, which is their explanation for why cardiac tasks under-generalize — the strongest mechanistic defence available on is-interoception-domain-general.
• The gain-term reading of accuracy. interoceptive-sensitivity records that if sensitivity is precision rather than signal fidelity, then Dunn’s result — accuracy as a pure multiplier with no main effect — is exactly what should be observed from the outside.

The psychophysical sense

A psychometric function has two parameters that matter: threshold (how big a signal you need) and slope (how sharply you go from chance to reliable as the signal grows). The first is sensitivity. The second is what the psychophysics literature calls precision — a shallow slope means inconsistent responding at a given stimulus magnitude, a steep one means the same stimulus produces the same judgment each time.

The HRDT and RRST estimate both. That is their point, and it is the reason Banellis et al. (2026) can say something about precision at all: as they note, the domain generality of this measure “has not yet been assessed” because no prior interoceptive instrument produced it. The heartbeat counting task yields a single error score with no slope in it.

The two senses are not the same thing, and this is the page’s main job

It is tempting — and the field is not always careful about this — to read a psychophysical slope as an estimate of the Bayesian precision term. They are related: a system that weights a channel’s errors highly should, other things equal, respond to that channel more consistently. But:

• A slope is a property of one task in one channel on one occasion. The theoretical term is a dynamically allocated weight that is supposed to shift with context, attention, arousal and expectation — the whole point of precision in active inference is that it is not a fixed trait.
• A shallow slope has many non-Bayesian causes: attentional lapses, motor error, criterion drift, fatigue, a noisy stimulus device.
• The theoretical term is defined over prediction errors relative to a model. Nothing in a threshold/slope fit identifies the model, so nothing identifies the errors.

So the honest statement is that the psychophysical slope is the first empirical handle on a construct the wiki has been explaining things with for years, not a measurement of it.

What the first cross-modal data say

Cardiac and respiratory precision are uncorrelated (r = −0.022, BF01 = 6.30). Two readings, both live:

1. Against a general precision trait. If precision-allocation were a stable person-level style — some people weight the body heavily, others discount it — it should show across channels. It does not. That is awkward for every wiki claim that treats “this person weights interoceptive signals strongly” as an explanation of anything, including the rubber-hand account above.
2. For precision as it is actually theorized. Precision is supposed to be allocated per-channel and per-context; a system that assigned one global weight to all bodily signals would be a badly designed one. On this reading the null is a confirmation, and Garfinkel’s differential-weighting proposal predicts it.

The uncomfortable part is that reading 2 saves the theory by making it very hard to test. If precision is channel- and context-specific by design, then no pattern of cross-channel correlation could count against it, and the construct’s explanatory reach — which is enormous on this wiki — is not matched by any constraint on what it forbids. That is the same falsifiability shape already recorded for degeneracy and theory-of-constructed-emotion, arriving in the computational lineage rather than the constructionist one.

What is missing

• Precision measured while it is manipulated. Attention, threat and arousal are all supposed to change precision-weighting. Nobody in the wiki’s literature has measured a slope before and after such a manipulation, which is the minimal test that a slope tracks the theoretical quantity.
• Precision related to any outcome. The wiki has accuracy↔outcome results (is-more-interoceptive-awareness-better) and nothing comparable for precision. The construct’s clinical claims are entirely theoretical so far.
• Precision on the efferent side. interoceptive-control and the sensory-control-loop require precision over action as well as perception; no instrument here touches it.