Consolidation
You can use consolidation to materialize delivery artifacts into durable graph context. Consolidation is how a spec decision becomes a future KG query result.
The consolidation pipeline
Queue entries are claimed, processed, and committed under graph writer controls so the embedded store is not written concurrently.
L1: DeterministicWorker
The deterministic worker extracts known structures: decisions, requirements, API contracts, test scenarios, business rules, ORNs, and parent links. It does not need an LLM.
L2: Cognitive agent
The cognitive layer can add alternatives, assumptions, learnings, and selected relationships. It runs after deterministic extraction and stays behind reconciliation.
Edges carry metadata such as layer, rule ID, creator, and fallback reason. This lets operators distinguish deterministic, cognitive, fallback, and legacy links.
Advisory lock
Pulse uses a per-board writer guard around graph commits. If a writer is busy, work retries rather than corrupting the graph.
Automatic vs manual consolidation
Most consolidation is automatic. Use manual sessions when an agent needs to stage candidate nodes and inspect reconciliation before committing.
Manual consolidation session
Abort a stuck session
Related pages
Last modified on May 8, 2026
