authored links in a directed graph.
explicit associations are formal, durable edges between memories. each edge has a relation type, a state, and a strength. these are first-class records—deliberately created, not computed.
associative memory
how memories find each other.
memories relate to each other. shellbrain captures these relationships in two ways: explicit associations that are deliberately authored, and implicit associations computed from semantic similarity. both are first-class. both expand retrieval past the first obvious match.
explicit associations.
strength is earned.
each edge accumulates observations over time. valence, salience, and optional problem context build strength gradually. associations are reinforced by repeated evidence, not assigned once.
how recall uses them.
recall walks explicit association edges up to two levels deep with a minimum strength threshold. these candidates land in the explicit related section alongside problem-attempt and fact-revision links.
implicit associations.
computed from semantic similarity.
implicit associations are not stored—they are recalculated during recall. vector embeddings determine which memories are conceptually close, regardless of whether anyone explicitly linked them.
bounded traversal.
recall walks semantic neighbors up to two hops via breadth-first traversal. not unlimited exploration. bounded depth, bounded breadth.
scored by decay.
explicit neighbors: anchor score times relation strength, divided by depth. implicit neighbors: anchor score times similarity, divided by hop. everything further from the direct match scores lower. distance costs.
recall that looks like creativity.
chaining through multiple associative hops surfaces memories the agent would never think to look for directly. every hop is explainable, every edge is inspectable, and every result says why it was included.