Working Memory Guides Perceptual Decisions Through Fast Capture and Slow Drift

Reviewer #1 (Public review):

Summary

This study examines how working memory (WM) influences perceptual decisions, with the aim of distinguishing fast attentional capture-like effects from slower, sustained perceptual biases. The authors use a dual-task design in which a perceptual estimation task is embedded within a WM delay, combined with a time-resolved analysis of mouse tracking reports and hierarchical Bayesian modeling. This approach reveals two temporally distinct signatures of WM-perception interactions within single trials, arguing against a unitary account of WM-driven perceptual bias and instead supporting multiple processes that operate over different timescales.

Strengths

A major strength of the study is its innovative use of a time-resolved mouse trajectory analysis to move beyond endpoint measures and reveal the dynamic evolution of decision biases. By decomposing trajectories into components that are and are not explained by the final response, the authors provide compelling evidence for an early transient deviation and a slower, endpoint-consistent drift. The combination of rigorous experimental design, hierarchical Bayesian modeling, and converging analyses yields compelling support for the central claims and offers a valuable framework for studying top-down influences on perception.

Weaknesses/points requiring clarification:

(1) The primary weakness concerns the clarity of the theoretical framing linking the identified trajectory components specifically to attentional capture and representational (or perceptual) shift. While the manuscript reviews prior work on attentional and perceptual biases, the conceptual transition to the proposed distinction between capture and representational shift would benefit from a stronger connection to the existing literature. Clarifying this relationship would strengthen the interpretation of the results.

(2) The use of the term "continuous" to describe the trajectory analyses may be confusing for readers, as it could be interpreted as referring to a continuous task rather than a time-resolved analysis of movements performed to make a discrete response.

(3) Figures 2 and 7 present posterior distributions of hierarchical Bayesian parameter estimates for endpoint responses in Experiments 1 and 2. However, they do not show how these model estimates relate to the raw behavioral data. Including model fits alongside the observed data would help readers assess the quality of the fits and better evaluate how well the modeling captures the underlying behavioral responses. Similarly, it would be helpful to see individual means in Figure 3a, panel 2, as is done in Figure 4.

Reviewer #2 (Public review):

Summary:

This manuscript investigates the mechanisms by which visual working memory (WM) interacts with perceptual judgements, using continuous mouse-tracking to dissociate putative attentional capture from representational shift. Across two experiments, participants maintained a color in WM while performing an intervening perceptual matching task. Analyses of mouse trajectories revealed bidirectional influences with distinct dynamics of attentional capture and representational shift components. For WM's influence on perceptual judgments, trajectories showed a fast and endpoint-inconsistent deviation (interpreted as attentional capture by WM-matching features), followed by a slower and sustained drift that closely matched the final perceptual bias. In contrast, when perceptual judgments influenced subsequent WM recall, trajectory dynamics were dominated by the sustained drift component, with minimal capture-like deviation. Together, these findings are interpreted as evidence that WM shapes perceptual decisions through at least two temporally distinct processes.

Strengths:

I find the paradigm to be cleverly designed and the analyses rigorous. A major strength of this work is the use of continuous mouse-tracking and time-resolved analyses to dissociate transient influences from sustained biases within single trials. The trajectory decomposition provides an elegant way to separate early deviations from later drift, which would be difficult to achieve using traditional measures that only measure the final recall. I find the observation particularly compelling that trajectories initially deviate toward WM-matching information and then correct back toward the task-relevant target, highlighting the dynamic interplay between transient priority signals and the final decision.

Weaknesses:

(1) The early curvature in the mouse trajectory, inconsistent with the endpoint, is interpreted as fast attentional capture. However, this signal may also reflect competition among multiple responses driven simultaneously by the WM representation and the perceptual matching item. While the current interpretation is plausible, it would be helpful if the authors could more clearly articulate why this component should be solely interpreted as attentional capture rather than early response competition.

(2) The mouse trajectories show a clear correction back toward the target later in the movement, particularly when the cursor enters the color wheel (Figure 3a), where the correction appears most pronounced. I wonder how this corrective phase should be interpreted. For example, does this correction reflect disengagement from an initial WM-driven priority signal, increasing influence of task demands and sensory evidence, or some other control process?

Relatedly, movement onset latency modulated the overall AUC but did not influence the final perceptual error. I wonder whether the time courses of the capture and shift components (as revealed by the destination-vector transformation) differ between early-onset and late-onset trials, and if so, when those differences emerge. Explicitly showing these comparisons would help further clarify how early capture is corrected while the endpoint bias remains stable. It may also be informative to include representative raw trajectory paths for early- and late-onset trials, as Figure 3a is currently the only figure showing raw trajectories, whereas most subsequent results are derived measures.

(3) The contrast in destination-vector dynamics between the perceptual matching response and the WM recall response (Figure 8) is interesting. For the representational shift component, the effect appears to increase sharply after movement onset. Conceptually, one might expect the shift in WM representation to have already occurred following perceptual judgment, rather than emerging during the response itself. It would be helpful if the authors could clarify why the shift is expressed primarily during the movement phase. Additionally, although weak, there appears to be a small capture-like deviation in the WM recall trajectories. Was this effect statistically significant? It may be informative to apply the same cluster-based permutation analysis directly comparing the capture effects against zero, in addition to the paired comparisons currently reported.