Prediction error, retrieval practice, and instruction–action congruence during memory reactivation.

The ability to adapt depends on updating predictions based on the discrepancy between expected and actual outcomes, known as Prediction Error (PE). Through reconsolidation, a PE-inducing reminder reactivates consolidated memories into an unstable state, allowing them to be restabilized and updated. We investigated which type of PE during reactivation—associated with task instructions, specific task cues, or their interaction—best enhances retention. Healthy adults (N = 102) learned 32 face–name pairs on Day 1. On Day 2, participants were assigned to a no-reactivation Control group or one of four reactivation conditions combining Instruction (Explicit vs. Non-explicit instruction) and Reminder type (Complete vs. Incomplete cue). Retention was evaluated on Day 3 via uncued recall and analyzed using generalized linear mixed models. All reactivation conditions significantly improved Day 3 retention relative to the non-reactivated Control group (p < .0001). Crucially, the condition combining two sources of prediction error, achieved the highest memory strengthening (p = .002). This advantage was most prominent at immediate testing (p < .001), before feedback-based relearning compressed group differences. These findings demonstrate that memory strengthening scales with the degree of PE induced during reactivation, providing direct support for a graded prediction-error account of memory reconsolidation.