When making decisions, whether as simple as determining if a color is more bluish or greenish, or as complex as choosing between career paths, there is always considerable variability in the confidence of our choices. This ability to evaluate our own decision-making processes and recognize potential errors in the absence of external feedback is known as metacognition (Flavell, 1979). Confidence judgements are often used to measure metacognitive ability, by matching the performance to a task with the level of confidence expressed afterward (Fleming et al., 2012; Fleming & Lau, 2014; Galvin et al., 2003). Research on metacognition has mainly focused on the perceptual domain, by asking participants to make a perceptual decision (e.g., judging the orientation or presence of a visual stimulus) and then report how confident they are in their decision (Fleming et al., 2012; Mamassian, 2016). In the present study, we investigate the role of motor information on confidence judgements.

Metacognition is not limited to perceptual decisions but extends to other domains, including motor decisions. To navigate efficiently in the world and adapt to changing environments, humans must evaluate the quality of their own actions. In our daily lives, we often receive direct feedback on the consequences of our actions, which helps guide future behavior. However, when this external feedback is absent or ambiguous, such as when navigating in the dark or grasping an object in a bag, we must rely on internally generated signals to evaluate our performance.

Several studies investigated sensorimotor metacognition by asking participants to rate their confidence during specific motor tasks. These studies have consistently shown that individuals have above-chance metacognitive ability in evaluating performance in tasks such as tracking a target (Fassold et al., 2023; Locke et al., 2020), throwing balls (Arbuzova et al., 2023), estimating limb position (Charles et al., 2020) or detecting errors in movement trajectory (Locke et al., 2020; Pereira et al., 2023). This suggests that individuals can compute reliable confidence judgements about their voluntary movements.

These findings prompt the crucial question of which informational sources individuals rely on to form sensorimotor confidence. Many theoretical models view metacognition as an inference process integrating multiple cues from both internal and external sources (Allen et al., 2016; Fleming et al., 2015; Kiani & Shadlen, 2009; Petrusic & Baranski, 2003). When computing sensorimotor confidence, several types of information might be available, such as prior knowledge about the motor system, efferent signals from the motor commands, proprioceptive afferent signals (e.g., about the position, strength or velocity of the limb) and perceptual feedback (e.g., seeing one’s hand moving) (Grünbaum & Christensen, 2024).

Some studies attempted to disentangle the contribution of these cues by comparing metacognitive performance across different conditions. For instance, one might assume that individuals have better metacognitive insight for voluntary movement, due to access to both efferent and afferent signals, which is not the case for passive or observed movements. However, Charles et al. (2020) compared metacognitive abilities across voluntary actions, passive movements, and visual signals. The results showed higher average confidence for voluntary movements but similar metacognitive ability across all conditions. This indicates that participants were not better at discriminating their correct from incorrect responses in the active condition than in the other conditions, suggesting that additional efferent information does not enhance metacognitive ability. Similarly, Arbuzova et al. (2023) compared metacognitive performance across motor, visuomotor and visual conditions of a ball-throwing task. Participants had to decide which one of two trajectories presented on the screen corresponded to their own movement and then rated their confidence. The ball trajectory was displayed on the screen in the visuomotor condition but not in the motor condition. In the visual condition, participants passively watched a replay of their movement. Results again showed no significant differences in metacognitive performance across the three conditions, suggesting that participants can accurately monitor their voluntary movement in the absence of visual feedback. Fassold et al. (2023) investigated the relative contributions of prospective and retrospective cues to sensorimotor confidence in a reaching task. Prospective cues involve information available before the action, such as prior knowledge, motor noise and past performance while retrospective cues include sensory feedback obtained after the action, like proprioceptive and visual feedback. The findings revealed that participants primarily relied on prospective cues to form their confidence judgements, while the extent to which individuals integrated retrospective cues varied. Some participants largely disregarded the sensory feedback that could have informed them about their actual performance.

While these studies provided valuable insights into how individuals integrate several internal and external cues to compute sensorimotor confidence, they overlooked the role of motor awareness in these processes. Research indeed suggests that the execution of voluntary movements often occurs without conscious awareness. Individuals typically focus more on the outcomes of their actions than on the ongoing adjustments made during the movement (Blakemore et al., 2002). For instance, a key study by Fourneret and Jeannerod (1998) demonstrated that individuals failed to consciously detect small deviations that were experimentally introduced into their movements. In their study, participants were asked to draw a straight line while their arm’s actual movement was hidden, and only manipulated visual feedback was provided. Despite adjusting their movements to compensate for the perturbation, participants were unable to verbally report these adjustments, indicating that they had only limited awareness of their proprioceptive feedback during the task. Similarly, other studies found that participants adjusted their muscle activity to compensate for perturbations without noticing the adjustments (Franklin & Wolpert, 2008; Maselli et al., 2022).

If individuals are unaware of the details of their movements, one might expect that their metacognitive judgments about these movements would also be limited. Recently, a study inspired by the work of Fourneret and Jeannerod examined whether unconscious deviations in movements impair confidence judgments (Pereira et al., 2023). Participants were asked to draw a straight line to a target, using a joystick, with visual feedback provided on a screen. A deviation was introduced in some trials, requiring them to adjust their movements. After each trial, participants were asked to report whether they noticed any deviation and to rate their confidence in this judgment. The results revealed that, although participants could not consciously report certain deviations, their confidence judgments still showed sensitivity to the presence or absence of these deviations. For instance, participants reported higher confidence for correct rejections (reporting no deviation when no deviation was present) compared to misses (failure to detect a deviation). This suggests that participants were able to make confidence judgments that reliably tracked objective performance, even when they failed to consciously detect the deviations. However, several outstanding questions remain unanswered. First, participants had access to visual feedback on their trajectory while performing the task, which means they could have relied on external visual cues to infer their confidence. This overlap makes it difficult to disentangle the contributions of internal motor signals from external feedback, leaving it unclear whether participants relied equally on their internal motor signals when they were aware or unaware of their movements. It also raises the question of whether the use of motor information for metacognitive judgments is contingent upon conscious detection of the movement. Second, the interplay between visual feedback and motor signals remains poorly understood. How do these sources of information complement or compete in shaping sensorimotor confidence? For example, if participants receive strong visual cues favouring a particular response, it is possible that they rely less on the internal motor information and vice versa. Third, the use of externally generated deviations does not accurately reflect the subtle, self-initiated corrections that are typical in real-world motor tasks. In everyday life, small adjustments in our movements are endogenous, arising naturally as we fine-tune our actions to achieve desired outcomes. This leads to a critical question: are individuals aware of these purely endogenous motor adjustments, and can they accurately assess their performance in the absence of external perturbations? Understanding this dynamic is crucial, as the ability to monitor and evaluate one’s own movement without relying on externally induced errors is likely to provide a more realistic understanding of motor control. Finally, none of the previous studies have directly measured the strength of motor signals using objective physiological recordings. This is an important limitation, as such measurement could provide more precise insight into the role of motor signals in sensorimotor confidence.

With these open questions as a guideline, the present study focuses on the role of purely endogenous motor corrections in the formation of sensorimotor confidence judgments. To achieve this, we used electromyographic (EMG) recordings of both hands during between-hand choice tasks to capture small motor activities called “Partial-Errors” (PEs). Partial-errors are subthreshold motor responses that occur on the incorrect hand just before the correct response is executed (Burle et al., 2002; Smid et al., 1990). On partial-error trials, participants thus nearly press the incorrect button but manage to correct themselves in time to select the correct one. These PEs reflect the operation of an online monitoring system that continuously evaluates the accuracy of motor commands and attempts to correct potential mistakes before they are fully executed (Meckler et al., 2017; Spieser et al., 2015). Despite being efficiently corrected, most PEs remain unconscious (Ficarella et al., 2019; Rochet et al., 2014). In the study of Rochet et al. (2014), participants could consciously report only about 30% of their PEs, with significant inter-variability in detection rates, ranging from 0% to 65%. This variability makes PEs an ideal phenomenon for studying the metacognition of subtle motor signals, whether they are consciously perceived or not. Rochet et al. (2014) identified two key parameters that predict the conscious detection of PEs: the magnitude, or the amplitude, of the PE, and the correction time (the interval between the incorrect and correct activations) (Fig. 1A). Larger PEs and those with longer correction times were more likely to be consciously detected. The amplitude of a PE reflects the intensity of the initial, erroneous motor activation, with larger amplitudes indicating a stronger motor impulse toward the incorrect response. Correction time represents the time the system takes to detect and correct the error. Together, these two parameters provide a measure of the motor signal’s strength and the efficiency of the corrective process. By examining the correlation between PE amplitude or correction times and confidence ratings, we can directly investigate how variations in motor signal strength influence participants' confidence in their own error detection. If stronger motor signals (i.e., larger amplitudes or longer correction times) correlate with higher confidence ratings, this would suggest that participants use the intensity and timing of their internal motor feedback to rate their sensorimotor confidence.

We conducted two experiments in which participants performed variants of a conflict task designed to induce response conflict in half of the trials through incongruency between task-relevant stimulus dimensions and distracting information. After each response, participants were asked to report whether they detected a PE and then rate their confidence in that report. Electromyographic recordings from both hands were used to measure the presence, timing and size of these PEs. Partial-errors predominantly occur in situations where response conflict is present, typically during incongruent trials. Consequently, the visual feedback indicating whether a trial was congruent or incongruent could inform participants about the likelihood of making a PE. To explore whether participants rely on such external feedback, and to understand how it interacts with motor information, we varied the strength of the external signal across experiments. In Experiment 1 and Experiment 2, participants performed a perceptual conflict task where they indicated the direction of a target arrow (left or right) while ignoring the direction of a prime arrow. To investigate the influence of perceptual evidence on sensorimotor confidence, we varied the amount of available visual information across experiments by using a supraliminal prime in Experiment 1 and a subliminal prime in Experiment 2.