Abstract
A core function of episodic memory is to distinguish between overlapping experiences by converting similar inputs into distinct, non-overlapping representations, a process termed pattern separation. While anatomical models emphasize the role of specific hippocampal subfields, particularly the dentate gyrus, CA3, and CA1, less is known about how these computations unfold over time and influence memory-based decisions. Here, we use source-localized magnetoencephalography and computational modeling to examine how theta oscillations from the hippocampus as a whole are related to evidence accumulation during mnemonic discrimination. Participants performed the Mnemonic Similarity Task, in which they classified Repeat, Lure, and Foil images as "Old," "Similar," or "New." Event-related spectral and source activity confirmed reliable hippocampal engagement during the task despite its anatomical depth. We fit a hierarchical Linear Ballistic Accumulator model to behavioral data, estimating trial-by-trial drift rates as a latent index of mnemonic evidence accumulation, and examined whether hippocampal theta power predicted these dynamics. Left hippocampal theta was negatively associated with drift toward "New" responses on lure trials, while right hippocampal theta was positively associated with drift toward "Similar" responses on foil trials. These effects suggest that hippocampal theta selectively indexes partial-match sensitivity, with consequences that are beneficial or costly depending on whether the stimulus has an encoded memory counterpart. However, direct comparisons between hemispheres did not yield credible differences. These findings offer preliminary evidence that trial-level hippocampal theta fluctuations are related to the dynamics of memory-guided decision-making and demonstrate the feasibility of linking deep-source MEG recordings to computational models of evidence accumulation.