Semantic memory

Semantic memories are the compressed, generalizable knowledge that the consolidation pipeline extracts from episodic experience. Where episodic memories record specific events (at 14:32 on Tuesday, this specific input produced this specific outcome), semantic memories express what can be learned from many such events (when this class of input occurs in this operational context, outcomes of this type tend to follow). The consolidation worker clusters episodic events by embedding similarity, identifies recurring patterns across the cluster, and writes a semantic memory record that distills the pattern without reference to any specific episode. This compression has two benefits: it reduces the volume of records the retrieval pipeline must search through, and it produces representations that generalize better than any individual episode - the semantic memory about a backpressure pattern is directly useful for reasoning about new backpressure situations even if none of the specific episodic memories that produced it are retrieved. Semantic memories accumulate their own retrieval priority and trust scores through the same reinforcement pipeline as episodic memories, and they can be critiqued and suppressed by the forgetting system if they prove inaccurate.