Sustained Attention to Response Task (SART)
Robertson, Manly, Andrade, Baddeley & Yiend (1997). A stream of digits arrives; you press for all of them except one rare target, on which you must do nothing. The task builds a habit of responding and then punishes it — commission errors are the measure.
Why an interoception wiki holds it
Because of what it isn’t. Nearly every outcome in this wiki’s contemplative literature is a questionnaire, and the standing worry across does-mindfulness-enhance-interoceptive-accuracy and is-more-interoceptive-awareness-better is that self-reported improvement after a training that teaches you to value bodily awareness is partly the training teaching you to endorse items.
The SART cannot be endorsed. In Eusebio et al. (2022) it is the only measure in an eight-questionnaire battery that distinguished the two yoga arms, while the self-report measures moved equally in both. That inversion — behaviour differentiates, self-report does not — is the reason this page exists.
It is also why the transfer claim is load-bearing. There is no overlap between attending to sensations in a yoga posture and withholding a keypress to the digit 3, beyond the abstract demand to keep track of the present moment. Improvement therefore cannot be practice effects on the trained material, which is the standard deflation of cognitive-training results.
The three-level readout
The task supports progressively finer analysis of the same trials, and in Eusebio et al. the three levels dissociated — which is the useful methodological lesson:
| level | measure | what it indexes | Eusebio et al. result |
|---|---|---|---|
| errors | no-go accuracy | inhibitory control / lapse rate | group difference (p = .006) |
| distribution | tau | frequency of extreme-slow responses = lapses, mind-wandering | no group difference (ps > .110) |
| sequence | post-error RT change | post-error slowing; adaptive re-engagement or ruminative disruption | no group difference (ps > .24) |
Accuracy improved without tau falling. Whatever changed, it was not the frequency of attentional lapses — so the paradigm’s own mechanistic indices declined to explain its headline effect. A page recommending the SART should recommend all three levels precisely because they can come apart like this.
A design detail worth copying
Because the trial preceding a no-go trial may itself be the trial following an earlier no-go, pre-target RT baselines are contaminated by prior post-target effects. Eusebio et al. analysed only trials preceded by at least three correct go trials. Any pre-/post-error RT analysis on a SART needs an equivalent guard, and papers reporting post-error slowing without one should be read with that in mind.