'Eye Know': Gaze reflects confidence in explicit predictions while relying on a distinct computational mechanism

Our decisions are naturally accompanied by a feeling of how confident we are in their correctness. While typically measured experimentally via self-reported judgments, decision confidence is typically expressed through implicit sensorimotor behavior. Combining VR-based probabilistic learning paradigms with computational modeling across one exploratory and two pre-registered experiments (total N = 115), we examined whether post-decision gaze dynamics reflects an implicit second-order evaluation of decisions. Gaze direction was consistently aligned with participants decisions about a targets upcoming location, and fluctuations in this alignment tracked the latent probability of the decision. Moreover, gaze exhibited computational hallmarks of confidence, reflecting a second-order ocular assessment of the decision. Directly comparing gaze-based confidence and explicit, reported confidence revealed a modest yet significant correlation, suggesting that the two are partially overlapping second-order processes. Notably, reported confidence showed higher metacognitive sensitivity and increased serial dependence across trials, while gaze-based confidence offered a more immediate, trial-specific readout of internal confidence. Importantly, we found that under increased uncertainty, gaze and decisions diverged more frequently, reflecting strategic "hedging" behavior. These findings suggest that post-decision gaze reflects an implicit, confidence-like behavior that partially overlaps with confidence judgments. Thus, the embodied interaction between decisions and ocular behavior supports flexible and adaptive behavior when learning and acting in a dynamic environment.