Gabriela Oancea¹, Luc Tremblay¹; ¹University of Toronto

Producing effective goal-directed behaviours requires judging the quality of our actions. In some tasks, natural feedback is insufficient to rate our actions, so we rely on an internal estimate of success. Such metacognitive judgements are critical for skill acquisition because the alignment between confidence and performance can help or hinder skill acquisition. Visual information plays a critical role in supporting metacognitive judgements about voluntary movements (Culot et al., 2025). Indeed, metacognitive ability can vary for judgements made on the basis of information from different stages of an upper-limb reach (i.e., start, middle, and end) (Oancea et al., 2025). Because work on the visual control of movement suggests that windows of visual feedback prior to peak limb velocity are particularly relevant to limb-target regulation mechanisms (Tremblay et al., 2017), this study focused on metacognitive judgements based on visual cues presented during upper-limb reaches. Participants (n = 22) performed goal-directed reaches and received a 20-ms window of vision when real-time limb velocity exceeded one of five selected criteria, which corresponded to 2% to 48% of movement time. After each trial, participants judged their endpoint bias (under- vs. overshoot) and rated their confidence in their judgement. Overall, participants showed above-chance metacognitive ability, as they were more confident on correct than incorrect trials. More importantly, although one would expect metacognitive ability to increase as the movement unfolded, there was a non-linear relationship between metacognitive ability and movement proportion. Additionally, participants were not always able to use the information from the endpoint judgement to accurately rate their confidence. We conclude that a brief window of vision is sufficient for accurate scaling of confidence ratings with endpoint judgement accuracy, but only at specific “optimal” windows that may be related to online visuomotor control mechanisms. Additionally, confidence is not always a direct read-out of endpoint judgement accuracy.