Thirty-two participants performed a predictive -inference task for four 120-trial runs during functional scanning. Runs alternated between the low-noise (SD=10) and high-noise (SD=25) conditions, with the order counterbalanced across participants.