were supported by DFG, German Research Foundation (project number 222641018 -SFB/TTR 135). Data, code and pre-registration protocols are available at https://osf.io/kyhu7/ (Experiment 1) and https://osf.io/sy342/ (Experiment 2). The data discussed in this article were first published in "The Confidence Database", https://osf.io/s46pr/ (Rahnev et al., 2020). We have no known conflict of interest to disclose. AbstractWe can make exquisitely precise movements without the apparent need for conscious monitoring.But can we monitor the low-level movement parameters when prompted? And what are the mechanisms that allow us to monitor our movements? To answer these questions, we designed a semi-virtual ball throwing task. On each trial, participants first threw a virtual ball by moving their arm (with or without visual feedback, or replayed from a previous trial) and then made a twoalternative forced choice on the resulting ball trajectory. They then rated their confidence in their decision. We measured metacognitive efficiency using meta-d'/d' and compared it between different informational domains of the first-order task (motor, visuomotor or visual information alone), as well as between two different versions of the task based on different parameters of the movement: proximal (position of the arm) or distal (resulting trajectory of the ball thrown).We found that participants were able to monitor their performance based on distal motor information as well as when proximal information was available. Their metacognitive efficiency was also equally high in conditions with different sources of information available. The analysis of correlations across participants revealed an unexpected result: while metacognitive efficiency correlated between informational domains (which would indicate domain-generality of metacognition), it did not correlate across the different parameters of movement. We discuss possible sources of this discrepancy and argue that specific first-order task demands may play a crucial role in our metacognitive ability and should be considered when making inferences about domain-generality based on correlations.