Neural dynamics of reconstructing knowledge structure from temporally disordered inputs

Humans possess a remarkable ability to integrate fragmented experiences into coherent internal models. However, the neural dynamics supporting structure reconstruction from temporally disordered inputs remain poorly understood. Here, we show that knowledge reconstruction follows a delayed, two-stage neural trajectory. Two independent cohorts learned complex physics concepts from scrambled video lectures during behavioral testing or functional magnetic resonance imaging. Learners receiving disordered inputs required sufficient information accumulation before their internal models matched the expert-defined structure. The posterior cingulate cortex showed stronger correspondence with knowledge structure during updating, whereas the precuneus showed more sustained correspondence after reorganization. Functional coupling between these two regions predicted immediate learning success for disordered input, but this success did not fully translate into robust delayed retention. These findings reveal a two-stage neural mechanism for transforming fragmented inputs into coherent understanding, highlighting both the adaptive capacity and potential subsequent cost of learning from fragmented materials. Using behavioral ratings and functional magnetic resonance imaging, this study tracks how learners dynamically reconstruct coherent knowledge from temporally disordered lectures. The posterior cingulate cortex supports structural updating, the precuneus shows later sustained representation, and their coupling predicts immediate learning success.