Research round three: memory, replay, and compositional reasoning
Our third research review covered twelve publications. Here is what they found about the brain, and where each piece of evidence stops.
Replay reorganises experience, it does not replay it
Human recordings show that the brain revisits past experience in an order set by learned relationships rather than the order things were seen. That is a finding about reorganisation, not about storage. It motivates holding evidence in a structure that can be re-read in a different order to the one it arrived in.
What gets revisited is prioritised by expected usefulness
Modelling work explains replay as access to the memories most likely to improve the next decision. The models are normative, not a complete neural implementation, so we treat this as a design intuition rather than a mechanism to copy.
Memory functions can come apart from planning
In one rodent study, replay did not predict the next choice at all. That negative result matters more to us than the positive ones: it argues against treating any single memory process as the thing that produces good reasoning.
Familiar parts, recombined
Intracranial work points to solutions being assembled from familiar elements rather than found whole. Compositional recombination is the one idea in this round that maps most directly onto how our specialists build a position out of separately sourced pieces.
The honest caveat
None of this establishes that our system works the way a brain works. Several entries were reviewed from the abstract only, and the studies involve small numbers of participants or animals. We use them as a source of hypotheses about structure, and we say so rather than claiming biological fidelity.