Real-time reorienting during episodic retrieval rescues memories from suboptimal arousal-attentional states

Successful knowledge expression varies with one's arousal-attentional (A-A) state at the moment just prior to attempting to remember. Yet, whether such states causally affect retrieval remains unclear. Here we introduce PRISM (Pupil-based Real-time Intervention System for Memory), a closed-loop paradigm that continuously monitors pupil size to track A-A states in real time and, during suboptimal states (SOS), delivers a visual reorienting cue in the moment before a retrieval attempt. In 67 participants performing a goal-directed source memory and item recognition task, we show that pupil-defined SOS just prior to retrieval impair both item recognition and source recollection relative to optimal state (OS) trials. Critically, delivering a brief visual reorienting trigger cue on SOS trials in real time rescues item memory, but not source memory, raising item memory to a level not credibly different from real-time OS trials and providing causal evidence that phasic attentional reorienting during A-A-defined suboptimal states facilitates memory access at retrieval. Offline spline-detrending analysis confirms that the A-A state-memory relationship is robust to slow physiological drift. It also shows that the trigger benefit was credible for marginally suboptimal (drift-driven) SOS trials and smaller and less certain for deeply suboptimal (drift-resistant) SOS trials, consistent with a zone-of-recoverability account, although the difference between the two SOS subtypes was not credible. Furthermore, among deeply suboptimal trials the A-A-boosting trigger cue improved memory for low-A-A SOS (i.e., pupillary constriction) but not high-A-A SOS (i.e., pupillary dilation) trials, consistent with the classic inverted-U relationship between arousal and performance, although the difference between the two directions was not credible. Together, these findings causally link phasic arousal-attentional reorienting at retrieval to memory rescue and suggest that the depth and direction of suboptimality shape whether such rescue succeeds, connecting the readiness-to-remember framework to the classic arousal-performance inverted-U relationship. Moreover, these findings provide a further foundation for adaptive closed-loop technologies that would tailor intervention to the depth and direction of momentary arousal-attentional states, facilitating memory expression in real time.