This chapter situates the present account of metacognition within the framework of predictive processing (also known as predictive coding and active inference). On broadly Bayesian views, like those offered by the predictive processing framework, brains minimize expected prediction error by assigning precision to predictions of sensory information. The chapter argues that this machinery can model metacognitive feelings without collapsing them into generic error signals. First, it clarifies how confidence operates at multiple levels, showing where conscious-access signals plausibly arise. Second, it diagnoses a “Bayesian triviality” pitfall: if all control reduces to precision-weighting, metacognition looks uninformative. The chapter avoids this by distinguishing subpersonal precision estimates from embodied, affect-laden, consciously accessible noetic feelings that re-parameterize control at the personal level and become available for social uptake. It then examines metacognitive biodiversity via comparative psychology: many species exhibit uncertainty monitoring and opt-out behaviors, yet fall short of normative self-knowledge. The missing ingredient is embedding: social practices that interpret, contest, and stabilize the meaning of felt confidence and doubt as reasons-responsive states. The upshot is a reconciled picture: predictive brains supply the computational substrate; embodied noetic feelings supply leverage over control; social embedding converts control into epistemic commitments. Thus, predictive processing, carefully specified, supports rather than trivializes the transition from confidence to normative self-regulation.