RUL-016 — Design the ontology before importing into any graph

Ruling

Before importing/ingesting nodes into a graph (Neo4j→Memgraph, source ingest, etc.), design/extend the canonical KG ontology FIRST — never dump raw nodes. "The user pushed back hard on a raw Neo4j→Memgraph mirror" (GIGO).

Evidence

• Global memory feedback-ontology-before-graph-import: raw import copies the mess — duplicate label trees (Class/CodeClass), Session keyed 3 ways (257k dupes), audit/packet churn — into a second graph. knowledge_graph/canonical_schema.py (SPEC-106 FR-001) is the anchor: canonical LABEL_*, REL_*, CANONICAL_MERGE_KEYS (one merge key per label).

Enforcement

• gate scripts/kg_schema.py canon
• gate scripts/sweep_test_pollution.py canonical

Verification

scripts/kg_schema.py check reports drift between the live graph and the canonical schema (CI-required per ADR-023); scripts/sweep_test_pollution.py purges nodes that violate the canonical id-marker/merge-key contract before test runs seed new ones, keeping the operational KG's ontology from silently drifting.