Intelligence plane

Ontology, identity, knowledge graph

Ground agents in a shared meaning layer. Every entity has a CEID for audit and a SID for ML feature use; every edge in the knowledge graph carries an evidence_ref; every retrieval returns a typed evidence bundle that the Critic can validate.

CEIDs and SIDs

CEIDs are stable, human-readable handles for audit and prompts. SIDs are compact discrete-token sequences for ML feature use (retrieval, ranking, recommendation). Different jobs, different identities.

Anti-pattern prevented

Split-brain definitions:two teams defining "active customer" differently. The Ontology enforces a single canonical meaning per version, with governed evolution.

Metrics that matter

Evidence coverage, retrieval precision/recall on golden sets, conflict rate by relationship type, replay determinism per snapshot pin.

Evidence-bound retrieval

Retrieval is GraphRAG over the typed property graph. Every hop carries an evidence_ref; every bundle is pinned to a snapshot_version and an ontology_version so replay is deterministic.

Evidence bundle (Knowledge Substrate output)

{
  "snapshot_version": "kg_2026_05_03_T0930",
  "ontology_version": "ont.support@4.2.0",
  "seed": { "ceid": "order:ord_881", "entity_type": "Order" },
  "hops": [
    { "hop": 1, "edge": "order_belongs_to_customer",
      "evidence_ref": "oms:db:orders/881#row_v3",
      "to": { "ceid": "customer:cus_77" }, "confidence": 0.99 },
    { "hop": 2, "edge": "customer_has_segment",
      "evidence_ref": "crm:exports/segments_2026_05_01.parquet#offset_4112",
      "to": { "value": "vip" }, "confidence": 0.96 }
  ],
  "conflicts": [],
  "truncated": false
}