The study investigates the role of dopamine in reward-guided decision-making, particularly in the context of hidden-state inference. Using a two-step task for mice, the researchers found that dopamine reports reward prediction errors (RPEs) using value information inferred from task structure knowledge, reward rates, and movement. However, neither activating nor inhibiting dopamine neurons at trial outcomes affected future choices, indicating that the influence of rewards on choices is not mediated by dopaminergic RPEs. Instead, these effects are driven by dopamine-independent information about the world's state. A neural network model was developed to simulate these findings, where the cortex infers hidden states by predicting observations, while the basal ganglia learns values and actions via RPEs. This model recapitulates the observed behavioral and neural activity patterns, suggesting that flexible behavior involves two parallel processes: state inference and value evaluation. The findings reconcile the roles of dopamine and hidden-state inference in reward-guided decision-making, providing a more integrated framework for understanding adaptive behavior.The study investigates the role of dopamine in reward-guided decision-making, particularly in the context of hidden-state inference. Using a two-step task for mice, the researchers found that dopamine reports reward prediction errors (RPEs) using value information inferred from task structure knowledge, reward rates, and movement. However, neither activating nor inhibiting dopamine neurons at trial outcomes affected future choices, indicating that the influence of rewards on choices is not mediated by dopaminergic RPEs. Instead, these effects are driven by dopamine-independent information about the world's state. A neural network model was developed to simulate these findings, where the cortex infers hidden states by predicting observations, while the basal ganglia learns values and actions via RPEs. This model recapitulates the observed behavioral and neural activity patterns, suggesting that flexible behavior involves two parallel processes: state inference and value evaluation. The findings reconcile the roles of dopamine and hidden-state inference in reward-guided decision-making, providing a more integrated framework for understanding adaptive behavior.