Saud Altayar¹ (), Keisuke Fukuda^1,2; ¹University of Toronto Mississauga, ²University of Toronto

Given that not all visual memories are accurate, we must distinguish accurate from inaccurate memories to guide our behaviour (Kozlova et al., 2025). To test whether such metacognitive judgments share a common mechanism between visual working memory (VWM) and long-term memory (LTM), participants completed a visual learning task in which they memorized either one, three, or five colored silhouettes of real-world objects at a time. Importantly, the color memory for half of the silhouettes was tested 1000 ms after encoding (assessing VWM) while the remainder was tested approximately 30 minutes after encoding (assessing VLTM). To quantify memory quality for each silhouette (i.e., target), participants used a continuous color wheel to indicate their best estimate of the target color (i.e., point estimate) and the range of all possible colors they thought the target could have been (i.e., certainty range). Participants’ memory distribution was captured as a triangle using the point estimate as the peak and the certainty range as the base. The peak’s height was set so that the triangle’s area was one. The cumulative area of the memory triangle was then computed as a function of the distance from the target color, and the area under this cumulative distribution function quantified the memory quality. To measure subjective assessment of the memory quality, participants also indicated their confidence in the point estimate’s accuracy. Finally, we built participant- and memory- specific regression models to explain the within-subject variability in confidence from the memory quality, and the resultant R-squared was used as a measure of metacognitive accuracy. A preliminary principal component analysis (n = 49/84) revealed a single latent factor that accounted for more than a half of the variance across set-sizes and memory types. This suggests a common mechanism used to assess the quality of visual memories across VWM and VLTM.

Acknowledgements: This research was supported by the Natural Sciences and Engineering Research Council (RGPIN-2024-05727).