Method at a glance
- Modality: cardiac
- Measures: Cardiac interoceptive sensitivity (threshold) and precision (slope) as separated psychophysical parameters, plus metacognitive bias and efficiency from trial-wise confidence — rather than a single accuracy score.
- Procedure: Five-second silent ‘heart listening’ phase with heart rate measured live by pulse oximetry; a train of tones is then played at a tempo offset from the participant’s current rate, and they judge faster/slower and rate confidence on a Guess-to-Certain VAS. Tone offsets are adjusted by the Psi adaptive algorithm to estimate the psychometric function in delta-BPM units; threshold (alpha), slope (beta) and lapse rate (lambda) are fitted, typically by hierarchical Bayesian modelling. A matched exteroceptive condition substitutes a reference tone train for the heart.
- Also called: HRDT, heart rate discrimination, Legrand task
Strengths
- Separates sensitivity from precision. The threshold says how far the tone must be from your heart rate for you to tell; the slope says how consistently you do it at a given distance. No counting-based measure yields the second, and Interoceptive precision is what the predictive-coding literature actually theorizes about.
- Yields a signed bias. The threshold is a deviation in beats per minute with a direction, so systematic over- or under-estimation of one’s own heart rate is a measured quantity rather than a confound — the exact thing unsigned counting error hides (see Is the heartbeat-counting task a valid measure of cardiac perception?).
- Carries a matched exteroceptive control by construction. The same faster/slower judgment can be made between two tone trains, so task-general effects (attention, criterion, scale use) can be subtracted. Almost no interoceptive method has this.
- Supports metacognition natively. Trial-wise confidence on a graded scale permits meta-d’/d’ (Metacognitive efficiency and metacognitive bias) rather than a post-hoc correlation between a questionnaire and a score.
- Adaptive staircase makes it efficient — usable estimates in 60–80 trials.
- Validated against the older cardiac tasks (Legrand et al. 2022: interoceptive threshold correlates with heartbeat counting accuracy, rs = .29), and open-source.
- The threshold is reasonably stable: test-retest r = .51 over about 47 days (Legrand et al. 2022).
Limitations
- The slope is not yet a reliable trait measure. Test-retest r = .10 in the validation study (Legrand et al. 2022), partly because the staircase changed between sessions. The slope is the parameter that makes this task the wiki’s handle on Interoceptive precision, so individual differences in it should be treated as unestablished.
- Exteroceptive references are not matched to the heart. Control tones were drawn from 40-100 BPM rather than around each person’s heart rate, so the two conditions sample different ranges (flagged by the authors).
- Intrinsically multisensory. The cardiac signal is judged against auditory tones, so cardiac performance is never measured in isolation. Banellis et al. (2026) flag this as a real asymmetry when comparing against a unimodal respiratory task, and propose a tactile-auditory control.
- Metacognitive estimates are fragile. In Banellis et al., 96 of 241 participants had to be excluded from cardiac M-Ratio analysis for negative or extreme values, against 10 for the respiratory task. Any cardiac metacognitive result is computed on a heavily filtered subsample.
- The cardiodynamic confound is not escaped. Nothing here measures stroke volume. A more forceful heart is easier to judge as well as easier to count; the task fixes counting’s response problems, not the signal-amplitude problem shared by every cardiac method. See Heartbeat detection task.
- Beliefs about heart rate can still enter. A participant who believes their resting rate is 70 has a usable non-interoceptive prior for a faster/slower judgment. The five-second listening window is meant to force a current estimate, but nothing verifies that it was used.
- The five-second window is a design choice (limiting heart-rate-variability noise) whose consequences are unexamined — the authors themselves ask whether participants maintain a rolling estimate and over what interval.
- Requires pulse oximetry and real-time signal processing; the device itself can supply tactile cues, which is why a soft-clip sensor and repositioning instructions are specified (Murphy et al. 2019 showed measurement device affects cardiac accuracy).
Connections
Introduced by Legrand et al. (2022), The heart rate discrimination task: A psychophysical method to estimate the accuracy and precision of interoceptive beliefs — the Aarhus group’s cardiac instrument, and the wiki’s first cardiac task built to psychophysical rather than clinical specifications. Met first through Banellis et al. (2026); the methods paper is now held first-hand as Legrand et al. (2022) (N = 223, two sessions), whose headline results are a 7-8.5 BPM average underestimation of one’s own heart rate, lower cardiac than tone precision and metacognitive efficiency, and a reliable threshold with an unreliable slope.
What problem it was built for
The wiki’s cardiac evidence rests almost entirely on two 1980s instruments — Schandry counting and Katkin/Whitehead discrimination — both catalogued with their defects on Heartbeat detection task. The structural complaint, made by the validity debate and independently by signal detection theory, is that a single accuracy number confounds sensitivity with response bias, and reports neither.
The HRDT’s design response is to make the judgment comparative and the scale continuous. Instead of “how many beats,” it asks “faster or slower than yours” against a stimulus the experimenter controls in known units. That converts the measurement into a standard psychophysical estimation problem, with the familiar apparatus: adaptive staircase, fitted psychometric function, threshold and slope estimated separately, lapse rate absorbing inattention.
Note the subtitle’s honesty: the accuracy and precision of interoceptive beliefs. The task does not claim to measure the afferent signal. It measures what the person’s estimate of their heart rate is, and how tightly held.
Why it matters to this wiki beyond the cardiac literature
It makes cardiac and respiratory interoception comparable for the first time. Paired with the RRST, it yields the same four parameters from the same estimation pipeline in a second organ — which is the precondition for Is interoception domain-general or organ-specific? being answerable at all. Every earlier cross-modal null could be blamed on instrument mismatch; this pairing removes that defence.
It measures the quantity the theory pages are about. Interoceptive precision appears throughout Predictive coding, Interoceptive inference, Active inference and computational psychosomatics as the load-bearing term — precision-weighting is what the models manipulate — while the wiki’s empirical pages had no measure of it. The slope parameter is a first, partial answer.
What it does not fix
The task solves a response problem, not a signal problem. Everything on Heartbeat detection task that turns on cardiodynamics — stroke volume predicting performance, exercise transiently manufacturing accurate perceivers, an older heart being a quieter one — applies here unchanged, because a bigger beat is easier to judge as well as easier to count. Nothing in the psychophysical apparatus addresses signal amplitude.
Nor does it settle whether the minority-validity thesis (Van der Does et al. 2000) survives. That claim is that the counting distribution is a mixture of one perceiving subpopulation and one confabulating majority. The HRDT’s continuous threshold in physical units is the right instrument to test it — a formal mixture analysis on HRDT thresholds in a large unselected sample is now feasible, and is the single missing analysis that debate has been asking for. The Banellis dataset (N = 513 for cardiac) is public and large enough.