Rhythmic replay of short-term memory neural patterns revealed by time-resolved error prediction

Theta oscillations are thought to provide a temporal scaffold for short-term memory (STM), organizing item encoding and maintenance into successive phases to reduce representational conflict. Whether this rhythm also determines encoding fidelity, that is, how precisely items are encoded in human cortical activity, remains unclear. Here, we show that STM encoding fidelity fluctuates rhythmically at theta frequency. EEG was recorded while participants encoded arrays of colored, oriented objects under two memory loads and, after a delay, reported the features of a retrospectively cued item on a continuous scale. Using time-resolved multivariate pattern analysis, we predicted subsequent recall error from encoding- and maintenance-period activity. Fronto-parietal theta- and beta-band activity predicted subsequent memory error. This prediction was not sustained but fluctuated at theta frequency and was not modulated by memory load. Cross-temporal generalization indicated that the same neural pattern recurred across encoding and maintenance, underlying the rhythmic fluctuations in memory-error prediction. Prediction fluctuations were temporally offset across spatial positions and object features. These findings characterize STM encoding as a rhythmic, recurrent process and link behavioral theta fluctuations to a distributed neural mechanism.