Memory layer

Memory graph

A visual way to inspect memory nodes, relationships, citations, and evidence.

Nodes

Nodes represent memory objects such as decisions, facts, constraints, risks, knowledge summaries, or evidence groups. A node should be inspectable so users can see where it came from and why it matters.

Edges

Edges connect related memory. Useful relations include SUPPORTS, CONTRADICTS, REFINES, and SAME_THEME. A graph with only SAME_THEME is useful for grouping but is less semantic.

Graph should be operational

The graph should represent the same structure retrieval uses. Runtime retrieval now expands through stored memory links, while the UI can also show inferred display links. If the graph shows isolated nodes, it usually means Conflux does not yet have enough stored relations, topic clusters, or semantic links to retrieve that branch confidently.

Current vs target

CurrentMemory records, stored links, evidence citations, compiled truth, graph-aware retrieval, and inferred topic links for display.

GapUI display links and runtime retrieval links are closer, but not fully unified as first-class root/topic/entity graph nodes.

TargetWorkspace root, topic clusters, evidence nodes, semantic edges, and C-to-B-to-A retrieval.

Quality signalOrphan rate, semantic edge rate, topic coverage, and retrieval trace should show whether memory is usable.

Evidence inspector

The inspector should show the selected memory, linked evidence, source user or client, affected files or artifacts, citations, and relationship context. This makes memory reviewable instead of opaque.