Sarah N. Garfinkel
The wiki had been citing Garfinkel’s distinctions since its taxonomy pages were built, always by name and never first-hand: the accuracy / sensibility / awareness three-way split (interoceptive-taxonomy, interoceptive-sensitivity) and the interoceptive trait prediction error (interoceptive-trait-prediction-error) are both hers, and both are load-bearing across the clinical and measurement pages. The Khalsa roadmap named her among its ~24 authors without giving her a page (it gave one to Critchley); Quadt, Critchley & Garfinkel (2018) is the first paper in raw/ she authors that the wiki reads directly, so the page is created now under the load-bearing-figure convention (as with Schandry, Katkin, Ehlers, Craske).
Knowing your own heart: the dissociation the field runs on
Garfinkel et al.’s (2015) tripartite framework is the single most-used measurement vocabulary in this wiki:
- Interoceptive accuracy — objective performance on a behavioural test (heartbeat tracking/discrimination).
- Interoceptive sensibility — subjective belief about one’s own interoceptive ability (questionnaire, or rated confidence).
- Interoceptive (metacognitive) awareness — the correspondence between the two: insight into how good one actually is.
Its power is that the three routinely come apart — someone can believe they are attuned to their body (high sensibility) while performing at chance (low accuracy), a mismatch invisible to any single “interoceptive awareness” score. Farb et al. (2015) push back on her use of the word “awareness” for the metacognitive quantity (they reserve it for reportability and propose “coherence” instead) — a genuine, recorded terminology conflict, not a disagreement about the underlying constructs. See interoceptive-taxonomy for all three nomenclatures side by side.
The ITPE: the discrepancy as a clinical quantity
Garfinkel’s second load-bearing idea is to give the accuracy–sensibility gap theoretical status. In autism she finds low accuracy paired with elevated sensibility — an enlarged interoceptive trait prediction error — and the ITPE, not accuracy alone, predicts trait anxiety across both autistic and control participants. This is what lets the review reconcile the mixed anxiety literature (accuracy is sometimes up, sometimes down; the discrepancy is the stable correlate), and it is the empirical anchor for the predictive-coding reading of anxiety as noisy interoceptive input amplified by mis-set self-referential priors (with Paulus & Stein).
The cardiac cycle as a probe
A separate strand: the Sussex group’s use of cardiac timing — whether a stimulus lands on systole (baroreceptors firing) or diastole — to show the visceral state is an input to cognition, not a backdrop. Fear stimuli at systole are processed differently; cardiac afferent signals modulate memory and even the expression of racial stereotypes (Azevedo et al. 2017). This is the second-dimension “influence” level of interoception made experimental.
The measurement problem, restated as a modelling problem (2021)
Her second first-hand appearance, Petzschner et al. (2021), is a useful complement to Quadt et al. because it gives a mechanistic reason why her tripartite split was necessary in the first place. The review’s “receiver characteristics” section notes that a heartbeat is not one signal: it is transduced across sensors separately encoding the occurrence of a pulsation, its strength, the associated pressures, and the chemical consequences — “quite distinct from vision or audition.” There is no unitary cardiac signal for a subject to be accurate about, which is one mechanistic reason accuracy and sensibility come apart and why the counting score behaves so oddly (is-the-heartbeat-counting-task-valid).
The same review makes her metacognitive dimension a formal requirement rather than a psychometric refinement: because metacognitive beliefs about one’s capacity to regulate the body influence both cognition and symptom development, “modeling an individual’s response to internal states and their appraisal may be as important as modeling the signals themselves.” That is interoceptive-taxonomy’s awareness dimension, arrived at from computational modelling instead of from questionnaires.
The cross-modal claim, and its failure to replicate
The fourth item in her contributions above — Garfinkel et al. (2016), Interoceptive dimensions across cardiac and respiratory axes — is the paper that licensed treating interoceptive awareness as a general capacity even while conceding that accuracy is channel-bound. It reported moderate cross-modal correlation in metacognitive insight (r ≈ 0.36) with first-order performance unrelated across the axes.
Banellis et al. (2026) tested that in 241 participants — powered to detect r ≥ 0.179, against her 49 — with psychophysical instruments designed for the purpose. The confidence half replicates and the insight half does not. Mean confidence correlates across cardiac and respiratory tasks (r = 0.51), and also with confidence about auditory tones (r = 0.61); metacognitive efficiency correlates across nothing.
Read through her own framework this is a sharper result than a simple non-replication. Her tripartite split exists to stop sensibility being mistaken for awareness. The new data say the cross-modal signal she found lives in the sensibility-flavoured component, and that this component is not specifically interoceptive at all. So the 2016 result is best read as her own distinction applied to her own finding — which is a use of the framework rather than a refutation of it. See is-interoception-domain-general.
Worth noting for balance: this leaves her the strongest surviving mechanism on the domain-general side. The differential-precision-weighting proposal in Quadt et al. — that channels differ in the weight assigned rather than in the ability available — predicts exactly the decorrelation Banellis et al. observe while preserving a common capacity underneath. It is also, as interoceptive-precision records, very hard to falsify.
The programme stated whole, in her own voice (Greenwood & Garfinkel 2025)
Everything above was assembled from her method papers and from others’ readings of them. Greenwood & Garfinkel (2025), her Annual Review of Psychology survey, is the first source in raw/ where she states the entire framework first-hand and points it at one target: emotion. Two things it settles for the wiki.
It gives the dimensional model its current canonical form. The tripartite split reconstructed above is here embedded in the eight-dimension Suksasilp & Garfinkel (2022) scheme (signal → neural representation → preconscious impact → accuracy → self-report/beliefs → insight → attention → attribution), each worked through an emotion example. So the “single measure is three” line above is really “is eight,” and the wiki now has her own statement of all of them rather than Quadt et al.’s summary. See interoceptive-taxonomy.
It is where her cardiac-cycle and ITPE strands become a clinical theory of emotion. The review runs the framework through six conditions — anxiety, depression, schizophrenia, PTSD, EUPD, autism — the last two new to interoceptive-psychopathology. It is also candid, in a way worth crediting to her: the Limitations concede the accuracy literature “has occurred in the cardiac domain, neglecting other bodily axes,” which is exactly the crack Banellis et al. widened into a domain-specificity result against her own 2016 cross-modal claim. Her framework, stated at its most confident, already names the assumption that undoes part of it.
Placement
Garfinkel anchors the Sussex/Brighton axis with hugo-critchley (her long-standing collaborator and co-author on nearly all of the above) and is allied with anil-seth on the predictive side of feedforward-vs-predictive-interoception. Where olga-pollatos and rainer-schandry represent the measure-accuracy-and-correlate-it tradition, Garfinkel’s contribution is to insist the single measure is three, and that the relationships among them — especially the accuracy/sensibility discrepancy — are where the clinical signal lives. Her caution about the heartbeat-tracking task’s generalizability (in Quadt et al.) puts her, unusually for a heavy user of the task, on the pro-caution side of is-the-heartbeat-counting-task-valid.