Perceptual glimpses are locally accumulated and globally maintained at distinct processing levels

Making decisions often requires the integration of multiple pieces of information. An extensive body of research has investigated the neural architecture underpinning evidence accumulation in perceptual tasks where information is continuously present, but less is known about how this neural architecture operates in situations affording only intermittent glimpses of an evidence source. In two electroencephalography (EEG) experiments, participants judged the direction of up to two pulses of motion evidence separated by gaps of varying duration. Our behavioural analysis found that participants used both pulses but underutilised the second, and showed no systematic decrease in accuracy as a function of gap duration. At the neural level, motor beta lateralisation tracked cumulative evidence across pulses, maintaining a sustained representation of the decision variable through the gap and until response. In contrast, the centroparietal positivity (CPP), a previously-characterised signature of evidence accumulation, built up transiently to a peak that scaled with each pulse’s contribution to the decision variable (i.e. the absolute belief update it produced), falling back to baseline in between pulses. These patterns were recapitulated in a model where pulse-information transiently integrated at the CPP level is fed to and maintained at a bounded motor level. In this model, the evidence in the second pulse is only integrated to the extent that the evidence in the first pulse falls short of the bound, or not integrated at all if a bound has already been hit.