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

Reviewer #1 (Public review):

Summary:

This paper characterises the physiological and computational underpinnings of the accumulation of intermittent glimpses of sensory evidence, with a focus on the centroparietal positivity and motor beta lateralization. The main finding is that the centroparietal positivity builds up during evidence accumulation but falls back to baseline during gaps, while motor beta lateralization maintains a continuous a sustained representation throughout the gap and until response.

Strengths:

- Elegant combination of electroencephalography and computational modelling.
- Innovative task design, including parametric manipulation of gap duration.
- The authors describe results of two separate experiments, with very similar results, in effect providing an internal replication.

Weaknesses:

- In their response to the reviewers, the authors now include a figure illustrating the relationship between the centroparietal positivity and motor beta lateralisation. However, in the absence of statistical analyses, it remains difficult to draw firm conclusions about this relationship.

- The paper does not provide an exhaustive characterisation across sensors and frequency bands. However, as the data are publicly available, these questions could be addressed in future work.

Reviewer #2 (Public review):

Summary:

This manuscript examines decision-making in a context where the information for the decision is not continuous, but separated by a short temporal gap. The authors use a standard motion direction discrimination task over two discrete dot motion pulses (but unlike previous experiments, fill the gaps in evidence with 0-coherence random dot motion of differently coloured dots). Previous studies using this task (Kiani et al., 2013; Tohidi-Moghaddam et al., 2019; Azizi et al., 2021; 2023) or other discrete sample stimuli (Cheadle et al., 2014; Wyart et al., 2015; Golmohamadian et al., 2025) have shown decision-makers to integrate evidence from multiple samples (although with some flexible weighting on each sample). In this experiment, decision-makers tended not to use the second motion pulse for their decision. This allows the separation of neural signatures of momentary decision-evidence samples from the accumulated decision-evidence. In this context, classic electroencephalography signatures of accumulated decision-evidence (central-parietal positivity) are shown to reflect the momentary decision-evidence samples.

Strengths:

The authors present an excellent analysis of the data in support of their findings. In terms of proportion correct, participants show poorer performance than predicted if assuming both evidence samples were integrated perfectly. A regression analysis suggested a weaker weight on the second pulse, and in line with this, the authors show an effect of the order of pulse strength that is reversed compared to previous studies: A stronger second pulse resulted in worse performance than a stronger first pulse (this is in line with the visual condition reported in Golmohamadian et al., 2025). The authors also show smaller changes in electrophysiological signatures of decision-making (central parietal positivity, and lateralised motor beta power) in response to the second pulse. The authors describe these findings with a computational model which allows for early decision-commitment, meaning the second pulse is ignored on the majority of trials. The model-predicted electrophysiological components describe the data well. Some flexible weighting of the second pulse also described the data well (in line with previous studies), but this explanation suffers from additional model complexity. In particular, this analysis of model-predicted electrophysiology is impressive in providing simple and clear predictions for understanding the data.

Weaknesses:

Behaviour in this experiment is different from previous experiments which use very similar designs (Kiani et al., 2013; Tohidi-Moghaddam et al., 2019; Azizi et al., 2021; 2023). The authors provide some possible explanations for this in the discussion. Overall performance in this experiment was much worse than previous experiments: Participants achieved ~85% correct following 400 ms of 33 - 45% coherent motion. In previous work, performance was ~90% correct following 240ms of 12.8% coherent motion. A second weakness is that, while bounded model can describe the data in this manuscript, it cannot explain the data from previous experiments showing a stronger weight on the second pulse.

Author response:

The following is the authors’ response to the previous reviews.

We have addressed the outstanding points made by the reviewers and provide a detailed description of the additional analyses performed & key results below. We have also updated the manuscript to reflect these additional results, and to contain a more detailed consideration of alternative plausible models.

We also note that we have corrected one figure panel (Fig.2 panel B, Exp. 2 only), where we identified a small bug in the visualisation code whereby the data of either one or two participants was not correctly plotted in some conditions. This makes no difference to the reported effects.

Please note that the reviewers acknowledged that your introduction now more broadly refers to the previous work from various groups on motor beta lateralisation (MBL).

(1) Evaluating the correlation between CPP and MBL, which is key for supporting the claim that CPP is feeding MBL. If, as you are alluding to in your rebuttal, single-trial estimates of CPP are too noisy, trials could be binned based on CPP.

As requested, we now provide additional analyses binning the data by CPP amplitudes, for the high-low coherence conditions at P1. Full details are provided below. In both experiments we find that, for a given coherence, greater CPP amplitudes at P1 correlate with stronger motor beta lateralisation.

(2) Examining the possibility of down-weighting (in line with previous studies) compared to your current bounded integration description. Specifically, does the your model predict a bi-modal CPP-P2 distribution that is not evident in the data?

We have now fit 3 additional models investigating alternative mechanisms that might account for the behavioural results. In particular, we have explored 4 different ways in which flexible weighting of the second pulse might account for both the behavioural and neural data. A full account of the results is provided below, and has been included in the manuscript. In sum, we find that a model which directly and uniformly downweighs evidence from the second pulse (as opposed to indirectly through little or no distance remaining to bound, as in our model) can account for the behavioural data well, but it cannot recapitulate CPP-P2 results unless an accumulation-terminating bound is also included in the model, and the additional complexity of a model with these two free parameters is not supported by model comparison. An alternative model where P2 is downweighed as an inverse function of P1 strength (i.e., stronger downweighing for P1-high coherence pulses) could recapitulate both the behavioural and neural data, but again, this was not favoured by model comparison metrics that account for complexity in the current dataset. We have added a piece on this in the discussion, noting how previous studies mentioned by the reviewer such as Cheadle et al., 2014 and Glickman et al., 2022 find a consistency bias where later evidence is boosted when it agrees with the earlier evidence, opposite to the dampening suggested by the model here, but that a key distinction in our task is that P2 always agreed with P1, so that a dampening might be plausible if subjects tend to withdraw some of their attention from the confirmatory P2 based on the strength of P1.

Regarding the CPP-P2 distribution, our original bounded model does indeed predict a bimodal CPP-P2 distribution with a peak at 0 arising from the early termination trials, which does not appear in our data (see Fig. S19 and related reply below). However, that EEG noise precludes the detection of any such bimodality in single trial amplitude distributions is demonstrated by the fact a bimodal distribution is strongly predicted for CPP-P1 amplitudes due to the two coherences, most strongly in fact for the unbounded model since there would be nothing to cap the higher-coherence, yet no trace of such bimodality is evident there either, due to EEG noise.

Public Reviews:

Reviewer #1 (Public review):

Summary:

This paper characterises the physiological and computational underpinnings of the accumulation of intermittent glimpses of sensory evidence, with a focus on the centroparietal positivity and motor beta lateralization. The main finding is that the centroparietal positivity builds up during evidence accumulation but falls back to baseline during gaps, while motor beta lateralization maintains a continuous a sustained representation throughout the gap and until response.

Strengths:

- Elegant combination of electroencephalography and computational modelling.

- Innovative task design, including parametric manipulation of gap duration.

- The authors describe results of two separate experiments, with very similar results, in effect providing an internal replication.

Weaknesses:

- A direct characterization of how the centroparietal positivity and motor beta lateralization interact is missing, which limits the novelty. In their reply to reviewers, the authors argue that the signal-to-noise ratio of EEG signals is insufficient for such analyses at the single-trial level. If so, a binned or trial-averaged approach could still be attempted.

As requested, we have now performed an additional analysis binning trials according to single-trial CPP-P1 amplitudes. To this aim, we sorted trials according to P1 coherence, and median-split them within condition according to the CPP-P1 amplitudes integrated in a time window around the grand-averaged peak [0.4 to 0.6s] after pulse onset, on the same subset of electrodes as in the manuscript. We then plotted motor beta lateralisation (MBL) as the difference in [Contra - Ipsi] hemispheres. Stronger negativities thus indicate stronger lateralisation towards the correct response. In all 4 cases, (both experiments and both coherence levels), higher CPP amplitudes were associated with stronger lateralisation from 0.5s post-pulse onwards (Author response image 1).

Author response image 1.

MBL (bottom) traces aligned to P1 onset (time = 0), median split by CPP amplitude [0.4-0.6s] post pulse onset, within P1 coherence condition. Trials with stronger CPPP1 potentials were linked to stronger MBL lateralisation toward the correct response.

- An exhaustive characterisation of sensors and frequency bands is also missing. In their reply to reviewers, the authors suggest that this would detract from their hypothesis-driven focus. I disagree: the main hypothesis and figures could remain centred on the centroparietal positivity and motor beta lateralization, with a more comprehensive mapping of sensors and frequencies placed in supplementary material. Since the purpose of the paper is to examine EEG-based decision signals in a novel behavioural context, a broader characterisation of the underlying EEG landscape would seem appropriate.

To broaden our characterisation, we have now included an additional supplementary figure that describes another distinct, relevant EEG signal. Fig. S12 shows the lateralised readiness potential (LRP), a lateralised motor preparation signal that has long been used as an index of relative motor preparation with high temporal resolution (Eimer, 1998; Kelly & O’Connell, 2013; Vidal et al., 2015).The LRP is typically computed as the difference in voltage between [IpsiContra] lateral motor electrodes with respect to eventual response, and it captures the fact that the contralateral motor cortex exhibits more pronounced negative ramps than the ipsilateral one immediately preceding action execution. The EEG landscape characterised in our paper thus comprises four distinct signals that are all functionally relevant to the task, including occipital alpha power, relevant for attention & temporal expectation encoding, which was included both in the main manuscript (Fig. 2) and the supplement (Figs. S9, S11). Given our already extensive supplementary material (18 figures) focused on our main research questions, we feel that a full, hypothesis-free exploration across the dimensions of frequency, space (sensors) and time, considering that there are 60 experimental conditions among which differences may be tested for (2 directions x 3 gaps x 4 coherence pairings in exp 1, plus 2 directions x 4 gaps x 4 coherence pairings in exp 2, plus single-pulse trials), would render the supplemental materials excessive in volume. Again, the data will be shared publicly for future exploration of these many dimensions.

Reviewer #2 (Public review):

Summary:

This manuscript examines decision-making in a context where the information for the decision is not continuous, but separated by a short temporal gap. The authors use a standard motion direction discrimination task over two discrete dot motion pulses (but unlike previous experiments, fill the gaps in evidence with 0-coherence random dot motion of differently coloured dots). Previous studies using this task (Kiani et al., 2013; Tohidi-Moghaddam et al., 2019; Azizi et al., 2021; 2023) or other discrete sample stimuli (Cheadle et al., 2014; Wyart et al., 2015; Golmohamadian et al., 2025) have shown decision-makers to integrate evidence from multiple samples (although with some flexible weighting on each sample). In this experiment, decision-makers tended not to use the second motion pulse for their decision. This allows the separation of neural signatures of momentary decision-evidence samples from the accumulated decision-evidence. In this context, classic electroencephalography signatures of accumulated decision-evidence (central-parietal positivity) are shown to reflect the momentary decision-evidence samples.

Strengths:

The authors present an excellent analysis of the data in support of their findings. In terms of proportion correct, participants show poorer performance than predicted if assuming both evidence samples were integrated perfectly. A regression analysis suggested a weaker weight on the second pulse, and in line with this, the authors show an effect of the order of pulse strength that is reversed compared to previous studies: A stronger second pulse resulted in worse performance than a stronger first pulse (this is in line with the visual condition reported in Golmohamadian et al., 2025). The authors also show smaller changes in electrophysiological signatures of decision-making (central parietal positivity, and lateralised motor beta power) in response to the second pulse. The authors describe these findings with a computational model which allows for early decision-commitment, meaning the second pulse is ignored on the majority of trials. The model-predicted electrophysiological components describe the data well. In particular, this analysis of model-predicted electrophysiology is impressive in providing simple and clear predictions for understanding the data.

Weaknesses:

Some readers may be left questioning why behaviour in this experiment is so different from previous experiments which use almost exactly the same design (Kiani et al., 2013; TohidiMoghaddam et al., 2019; Azizi et al., 2021; 2023). Overall performance in this experiment was much worse than previous experiments: Participants achieved ~85% correct following 400 ms of 33 - 45% coherent motion. In previous work, performance was ~90% correct following 240ms of 12.8% coherent motion. A second weakness is that, while the authors present a model which describes the data based on pre-mature decision-commitment, they do not examine explanations from the existing literature, that evidence is flexibly weighted, and do not provide any analyses which could be used to compare these descriptions. While their model can describe the data in this manuscript, it cannot explain the data from previous experiments showing a stronger weight on the second pulse.

The revised version of the manuscript includes a detailed discussion about possible reasons why our stimulus characteristics, task design & experimental protocol may have led to the observed behavioural results (lines 605 onwards). Furthermore, we have now included an extended model comparison as a supplementary note which examines alternative models that could account for the observed data and an additional discussion section that links it to the existing literature.

Recommendations for the authors:

Reviewer #2 (Recommendations for the authors):

The authors have responded to each of the comments in the previous review. The manuscript introduction and discussion have been substantially improved, and now more adequately address the previous literature. Limited improvements were made to the analysis, although the authors acknowledged why the suggested improvements from the reviewers were unlikely to be successful, but did not attempt to address the comments using other methods.

One common theme to both reviews was that, although the model broadly describes the data, it is not fully tested, and alternative descriptions are not fully considered.

We thank the reviewer for their careful consideration of our data & their reply. We agree it is important to formally test alternative descriptions, most particularly those involving downweighting of the processing of P2, and we have now done so. If participants were simply downweighting P2 by implementing generally smaller drift rates regardless of P1, we would expect the CPP-P2 to exhibit the same classical pattern as in P1, with higher amplitudes following high-coherence P2. The interaction pattern we observe, whereby CPP-P2 amplitudes are systematically lower following high-coherence P1, within each P2 coherence, can only be explained by some dependency between the processes occurring at P1, and those following at P2. What if, as the reviewer suggested originally, “additional evidence from the second pulse was down-weighted according to certainty following the first pulse?” We thus consider this also.

To formally test this, we have now fit four further models and compare both the fit quality to accuracy data and the predicted EEG results to our original implementation. All models were fit to the grand-averaged accuracy data for all conditions, using 10K simulations, and for both experiments separately (as was done for the bounded model presented in the manuscript). For EEG simulations in unbounded models, we assumed that the CPP signal fell back down to zero upon the dots turning blue, as for the simulations in the main manuscript. Fig. S15 illustrates the fit quality as measured by means of Bayesian Information Criterion (BIC), and we go into more detail on each model in turn below.

“(1) Unbounded model + P1-independent P2 drift rate reweighting (DR_P2)

First, we fit a model with no bound in which the drift rate for P2 was estimated by reweighting (positive or negative) the P1 drift rate via an additional free scaling parameter w. This model thus had the same complexity as our original one (k = 3 free parameters): two drift rate parameters for high and low coherence of P1 (d_high, d_low), plus a scaling parameter, w, dictating the strength of P2 relative to same-coherence P1. The drift rate for P2 (d_P2) was computed as the d*w, where d equals d_high or d_low depending on P2 coherence, multiplied by the scaling parameter w, so that if w < 1, P2 drift rates would decrease compared to the same coherence in P1. This model implements the reviewers’ suggestion that systematic downweighting of the second pulse might also account for the results (while also allowing for upweighting (w > 1) for the sake of flexibility).

This first model (DR_P21) yielded a similar fit quality to the behavioural data compared to our original bounded model (Bnd), as measured by BIC (Fig. S15). The new model broadly recapitulated the key behavioural results, including order effects, (Fig. S16,A) and could also recapitulate the generally lower CPP-P2 amplitudes (Fig. S16,B). However, this model failed to capture the key coherence-based pattern observed in the CPP-P2 data. Namely, while our EEG results showed that CPP-P2 in trials following P1-low coherence pulses reached overall higher amplitudes than that in trials following P1-high coherence pulses (see manuscript Fig. 3), this model’s simulations predicted that CPP-P2 should scale only with P2 coherence, showing higher amplitudes and steeper build-up rates for P2-high trials, regardless of P1 coherence (Fig. S16C). This is at odds with our empirical results.”

“(2) Unbounded model + P1-dependent P2 Drift rate reweighting (invDR_P2)

Next, we tested a model in which P2 downweighting could depend on P1 strength. That is, we made the P2 drift rate scaling parameter w inversely proportional to P1 coherence so that P2 drift rate d_P2 = d*w/d_P1, where d equals d_high or d_low depending on P2 coherence, and d_P1 indicates the preceding P1 coherence. This implements a kind of certainty weighting, whereby evidence following a strong P1 is more strongly dampened than evidence following a weak P1. This model could recapitulate the key behavioural findings (Fig. S17A), and also qualitatively captured the CPP-P2 effects (i.e. P1-low trials reaching overall higher amplitudes than P1-high trials, Fig. S17C), although the magnitude of this effect was substantially smaller than predicted by the original simple bounded model with no drift rate modulations. However, BICs indicated that this model provided an overall worse fit to the accuracy data compared to the original bounded model (Fig. S15).”

(3) Bounded models + P2 drift rate reweighting (Bnd + DR_P2, Bnd+ invDR_P2)

Finally, we investigated how well models with both a bound and either of the two P2 drift rate scaling methods we investigated above (uniform downweighting, P1-dependent downweighting) could capture the data, thus effectively testing two extensions of our original implementation that allowed for flexible reweighting of P2.

The systematic reweighting model with a bound (Bnd + DR_P2) could recapitulate all key behavioural and EEG findings (Fig. S18), but the additional complexity of the model was not supported by BIC (Fig. S15). Crucially, the reason that this model could recapitulate the CPPP2 results was still the presence of a bound, although we note that the proportion of trials that were predicted to terminate early was reduced in this model compared to the original bounded model presented in the manuscript (c.f. Fig. 4). Yet, this relatively small fraction of trials where accumulation ended early meant that 1) in some trials no accumulation was allowed to occur at all during P2, and 2) where it occurred, the DV was closer to the bound following P1-high coherence trials, thus needing to accumulate less further evidence before reaching a bound and yielding smaller CPP-P2 amplitudes overall in those trials. The inclusion of a bound was thus key to allow a model with systematic P2 downweighting to account for both behavioural and EEG data.

The model with P1-based scaling of P2 drift rates with a bound could also recapitulate all key behavioural and EEG findings (Fig. S18), but again, the additional complexity of the model was not supported by BICs (Fig. S15).

“Conclusion

This extended modelling exercise suggests that 1) a simple bounded model is favoured by model comparison, 2) an alternative model of equal complexity which includes a P1dependent systematic downweighting of P2 rather than a bound can produce qualitatively similar results, the common feature of both viable models being the push-pull relationship between P1 and P2, and 3) more complex models including both a bound and P2 modulations can also account for both the behavioural and EEG results, but the additional complexity is not supported by the current data. While it is possible that some flexible weight modulations occur, these are not sufficiently influential to justify its inclusion in the model. Future work would nevertheless be warranted to explore this possibility in more detail using tailored task paradigms. For the scope of this paper, we have maintained the bounded account in the main manuscript as it is the one supported by the model comparison in the current dataset, but we have also included the alternative P1-based downweighting account as a supplementary figure, along with some additional discussion.”

In response to my previous comment 3, the authors show their model predicts that there should be no CPP-P2 if the bound is reached before P2, otherwise CPP-P2 is similar to CPPP1 (Figure R2). The argument in the manuscript is that the lower CPP-P2 is because of this bound. The distribution of CPP-P2 amplitudes should therefore have higher variance than CPP-P1 amplitudes, and one might even predict a second mode in the distribution, around 0 amplitude (those trials that terminated before P2). The authors do not show this.

In Figure R3, it looks like the data have been normalised independently for CPP-P1 and CPPP2 (since the means are approximately the same); normalisation also prevents a comparison of the variance. However, it is apparent that there is no bimodality in the CPP-P2 distribution - were there substantially more trials with 0 CPP-P2 amplitude than CPP-P1? Is the EEG data actually more consistent with a model that systematically downweights P2?

In the previous Figure R3, data were normalised across CPP-P1 and CPP-P2, not separately. We plot the non-normalised values here (Author response image 2), for comparison, along with median, variance and skewness values for CPP-P1 and CPP-P2. Additionally, we attach the single-participant plots at the bottom of this document (Author response image 3)

We reanalysed the non-normalised data, excluding outliers (defined as values exceeding the mean +/- 3 times the standard deviation, computed for each pulse & for each participant separately). We found, in both experiments, lower median amplitudes (Exp. 1: t(21) = 2.68, p = 0.013; Exp. 2: t(20) = 4.37, p < 0.001), higher variances (Exp. 1: t(21) = 0.59, p = 0.55; Exp. 2: t(20) = 2.33, p = 0.03) and more positive skewness (Exp. 1: t(21) = 1.79, p = 0.08; skewness: 0.027 vs. 0.69; Exp. 2: t(20) = 3.85, p <0.001; skewness: 0.027 vs. 0.214) in CPP-P2 compared to CPP-P1, although variance and skewness effects were only significant in Exp. 2.

The reviewer argued in the previous review as well as here that increased variance would be predicted by the model – this is correct, but we believe that the mere presence of higher CPPP2 variances in our empirical data does not, on its own, necessarily support the model. That is because this increased variance could be explained by other factors, such as the increased EEG signal complexity of data at P2 compared to P1. This increased complexity naturally arises from overlapping potentials from CPP-P1 (which can be corrected for, but will increase data noise and thus variability nonetheless), as well as the various gap durations across conditions, which would also affect pre-P2 dynamics. Thus, while our data (partially - in Exp. 2 only) support the reviewer’s interpretation, we would be cautious in using the observation of increased variability as evidence for or against our model given the considerations above.

Author response image 2.

A. Empirical CPP–P1 and CPP-P2 amplitude [450-550ms post-pulse] distributions, pooled across coherences. Data were not normalised within-participant. Data were baselined 100ms before pulse onset prior to CPP-P2 amplitude extraction. In both experiments, CPPP2 amplitudes had a lower median (vertical line) amplitude and higher variance than CPP-P1. B. CPP amplitude simulations based on the original bounded model. A high number of trials where CPP-P2 amplitude should equal 0 due to early terminations. C. CPP amplitude simulations based on the inverse weighting model (invDR_P2). The model did not predict any trials with zero CPP-P2 amplitude because early terminations were not allowed. Rather, the mean distribution shifted towards lower predicted amplitudes, because P2 was downweighted proportionally to P1 coherence.

The reviewer asks: “Were there substantially more trials with 0 CPP-P2 amplitude than CPPP1?”. Given the noisiness of single-trial EEG data due to high-frequency artifacts, and/or spurious signal drifts (as illustrated by the raw CPP value distributions in Author response image 2A above), it is not be possible to directly detect trials on which CPP = 0. Instead, this must be inferred by other means. If the CPP is in reality at 0 on a larger proportion of trials, then on average, this should manifest as a higher fraction of trials with lower amplitudes, resulting in a more positively skewed distribution of CPP-P2 amplitudes compared to CPP-P1. As reported above, in both experiments we find that skewness is higher in CPP-P2 than in CPP-P1, in line with this hypothesis.

Regarding the question “Is the EEG data actually more consistent with a model that systematically downweights P2?”, we point to the additional modelling we conducted in response to the comment above. To recapitulate, we find that a model that systematically downweights P2 could account for behavioural findings, but could only recapitulate the key EEG CPP-P2 patterns if an accumulation-ending bound was also included in the model. Instead, a model where P2 is downweighted as a function of P1 strength could qualitatively capture both behavioural and EEG findings, but was not strongly supported by goodness of fit measures in this dataset. We note, however, that the latter model does not predict a bimodal CPP-P2 distribution, but rather a shifted mean and more positive skewness for CPP-P2 trials (Author response image 2C; skewness: CPP-P1 = 1.08; CPP-P2 = 1.24). Thus, in that respect, it does appear to provide a better qualitative recapitulation of the single-trial CPP-P2 data. However, given the extent of EEG noise, the bimodal underlying distribution of our bounded model would also translate to a unimodal, skewed distribution as observed, so this does not provide a strong basis for adjudication. Underscoring this, it is noteworthy that an unbounded model in fact predicts a more separated bimodal distribution for P1 than a bounded model, yet, again, with EEG noise, we are not able to identify any such bimodality in the empirical P1 amplitude distribution.

Minor:

In the discussion, the authors write "We also used a narrower range of coherences than the previous studies, which possibly lends itself to calibrating a bound to achieve acceptable accuracy while saving cognitive effort." (Page 22). Perhaps this should be reworded. The range of coherence in this study was ~26-44% in Exp 1 (a difference of 18%) in previous experiments the coherence was 3.2-12.8% (a difference of ~10%). The ranges of performance were similar.

We agree that the phrasing could be improved. We meant to say that we used only two coherences (high-low) with less than a twofold difference between them, instead of multiple levels of evidence strength in previous studies (e.g. 0,3.2,6.4,12.8) – we have clarified this.

The authors mention in their rebuttal "However, in contrast to previous studies, we did not include any feedback on a trial-by- trial basis, instead only providing feedback at the end of each block indicating the average accuracy." Actually, Kiani et al., 2013 also only gave feedback at the end of each block. This is also implied in the discussion. I suggest this be removed as the common feedback in Kiani et al., 2013 suggests this cannot explain the difference.

The methods in Kiani et al. 2013 state “At the end of motion stimulus, a 400–1000 ms delay period (truncated exponential) was imposed before the Go signal, disappearance of the fixation point, was presented. The subject was required to report the net direction of motion within 1 s after the Go signal by pressing a left or right key. Distinctive auditory feedback was delivered for correct and error responses. On trials with 0% coherence, the type of feedback was chosen randomly.“ We understand this means feedback was provided after every trial. If this interpretation is wrong, we would like to kindly ask the reviewer to point us to the relevant methods section so that we can correct the manuscript.

Author response image 3.

Individual CPP-P1 (blue) and CPP-P2 (orange), for both experiments (non-z-scored). Vertical lines indicate median CPP amplitudes for each pulse, respectively.