Allen (2026) — respyra: a respiratory tracking toolbox
The wiki’s first instrument paper, and its first source in which the interoceptive quantity being measured is not a perception at all but a control performance.
respiratory-interoception was created three ingests ago and closed with a complaint: breathing is objectively and continuously measurable, manipulable in a graded way, and yet “remains a secondary measure in a field bottlenecked on cardiac tasks,” with “none of them yet represented in this wiki by a primary source.” This is that primary source, and it arrives having read the same field the same way.
The gap it names
Interoception research has focused primarily on perception, with comparatively little attention to the control of bodily state.
That sentence is the paper’s whole reason for existing, and it is a fair charge against nearly every method page in this wiki. Counting and discriminating heartbeats, rating one’s bodily attentiveness, associating properties with emotion words, colouring felt topography — all of them ask what the person perceives. Even the field’s taxonomies, which are otherwise exhaustive to the point of redundancy (see interoceptive-taxonomy for four of them), have exactly one construct on the control side — Farb et al.’s regulation — and no task that measures it. The regulation slot has been vacant since 2015.
respyra is a candidate occupant. See interoceptive-control for what filling that slot commits the field to.
Why breathing, and why that is both the strength and the catch
Respiration is chosen for exactly the property the wiki flagged as making it unrepresentative: it is the one channel open to immediate voluntary control, so it is the only one on which a control task can be built at all. You cannot ask a participant to track a sinusoidal target with their gastric motility.
This cuts both ways, and the two pages should be read together. respiratory-interoception argues that respiration’s exceptionality is a problem for generalizing the field’s intervention evidence. Here the same exceptionality is a precondition for measurement: the control question can only be posed where control exists. So the wiki now holds both — respiration is a bad representative channel and the only tractable one — and the honest position is that interoceptive control may simply not be a general construct.
The borrowed paradigm
The methodological move is to import visuomotor adaptation wholesale from reaching research (Krakauer et al. 2019). Display the participant’s breathing trace, scale its deviation from centre by a gain g, leave the target where it is:
f_display = c + g · (f_actual − c)
At g = 1 the display is veridical and visual error equals physical error. At g = 2 every correction produces twice the expected visual displacement — the loop gain of the visuomotor control system doubles, corrections overshoot, and the stability margin drops. The participant must breathe with smaller amplitude to hit the same visual target.
Two conditions, two constructs, and the paper is careful that they are dissociable rather than harder-and-easier versions of one thing:
- Veridical MAE — baseline respiratory control ability.
- Perturbation ratio (perturbed/veridical) — the cost of remapping, normalized against baseline ability.
The validation data support the dissociation in the only way N = 1 can: fatigue hit the ratio (2.61 → 1.68 → 1.79 → 2.42, U-shaped) while leaving veridical performance flat. Whatever degraded across sessions was the remapping, not the breathing.
The RMSE/MAE ratio is the quietly useful part
Mean error says how big; RMSE/MAE says what shape. For normally distributed error the ratio is √(π/2) ≈ 1.253; above that means heavy tails — intermittent large deviations rather than a uniform increase in difficulty. Veridical trials sat at 1.25 exactly. Perturbed trials ran at 1.50, and Session 4 reached 1.72 with excess kurtosis of 10.6 and individual errors up to 10 N.
This is the metric that found the breath-phase effect. Inspiratory and expiratory MAE differed only marginally (p = .099), but their error shapes differed reliably under perturbation (p = .002) — inspiration, the active diaphragmatic phase, is more vulnerable to control instability than passive expiratory recoil. A distributional statistic recovering an effect the mean missed is worth noting in a literature that reports almost nothing but means.
Reliability, and the comparison it invites
Spearman–Brown corrected reliability of .857 at the trial level, from 48 trials in one participant. Set that against what the wiki holds on its flagship cardiac instrument: under 10% of healthy adults can do the heartbeat-counting task while 95% produce a number anyway (van-der-does-2000-heartbeat-perception-reanalysis), with the score confoundable by stroke volume and arousal, and the validity debate still unresolved decades on.
The comparison is not fair — one participant, one sitting, and a reliability estimate is not a validity argument — but the structural difference is real and is the toolbox’s best claim. A tracking error is a physical quantity measured continuously against a known target. It does not require the participant to report anything, so it cannot be a report of a belief about the body rather than the body. That is precisely the failure mode the cardiac literature has spent thirty years unable to rule out.
What it buys at that price: the thing being measured is no longer perception. Good tracking could reflect excellent respiratory interoception, or excellent motor control with the interoceptive contribution doing nothing. Nothing in this design separates them — the visual feedback is exteroceptive, and a participant could in principle track well on vision alone. The paper’s proposed remedy is correlational (correlate the perturbation ratio against respiratory perceptual sensitivity measures, per Nikolova et al. 2022), which would establish covariation, not that interoception is in the loop. Flagged here rather than on the method page because it is the central open question about the whole enterprise.
Forward pointers
The paper’s own proposals, recorded because the raw tree can test some of them:
- Correlate the perturbation ratio with respiratory perceptual sensitivity (Respiratory Resistance Sensitivity Task, Nikolova et al. 2022; Filter Detection Task, Harrison et al. 2021) to ask whether perceiving the breath predicts remapping it.
- Clinical populations with dysfunctional breathing — panic disorder, chronic hyperventilation, COPD — as selective-deficit tests. This connects directly to cognitive-model-of-panic and anxiety-sensitivity: if panic involves catastrophic misreading of respiratory sensation, the question of whether panickers also control the breath differently has never been asked with an instrument.
- Quantify how accurately people actually implement prescribed breathing patterns in breathing-based interventions — a fidelity measure for slow-breathing, which the wiki currently holds with no way to check whether participants did what the pacer told them.
Two adjacent papers in the queue
raw/papers/ holds two more from this group, both unprocessed at ingest time and both cited here:
- Banellis et al. (2026), Communications Psychology — now read. Interoceptive ability is uncorrelated across respiratory and cardiac axes: sensitivity, precision and metacognitive efficiency all null at N = 241, with moderate Bayesian support. The strong form of the channel-specificity worry, confirmed on matched instruments — see is-interoception-domain-general.
- Harrison et al. (2021), Neuron — interoception of breathing and its relationship with anxiety, the Filter Detection Task source. Still unprocessed.
What the Banellis result does to this paper. Two things, pulling opposite ways.
It strengthens the case for this instrument’s programme: if there is no general interoceptive ability, then the field needs good measures per channel rather than one better general measure, which is precisely the tooling stance recorded on micah-allen. Building a respiratory instrument stops looking like a substitute for the cardiac one and starts looking like the only coherent move.
It sharpens this page’s central caveat rather than relieving it. The worry recorded on interoceptive-control — that control measured through the breath may be a fact about breathing rather than a general capacity — was awaiting Banellis et al. as the test. It is not one. That study decorrelates perception across channels, which makes a general control construct less likely by analogy while testing nothing about control, and no second drivable channel exists to test it in. The question is now not merely open but unaskable with current instruments.