PageRank returning all zeros despite nodes existing is almost always a graph connectivity issue — PageRank requires edges to propagate scores. Here's the diagnostic path: ## 1. Check edge count ```cypher MATCH ()-[r]->() RETURN type(r), count(r) ORDER BY count(r) DESC ``` If this returns zero or very few edges, that's your problem. Nodes without edges all get the same base score (effectively zero after normalization). ## 2. Check if edges were created during ETL The ETL might be creating nodes but failing silently on relationship creation. Common causes: - Node property mismatch in MERGE clauses (creating duplicate nodes instead of connecting to existing ones) - Edge creation running before nodes are committed (transaction ordering) - UNWIND on empty arrays silently producing zero edges ## 3. GDS graph projection might be filtering edges ```cypher CALL gds.graph.list() YIELD graphName, nodeCount, relationshipCount ``` If `relationshipCount` is 0 but your actual graph has edges, the projection filter is wrong. Check your `gds.graph.project()` call — it needs to explicitly include your relationship types: ```cypher CALL gds.graph.project( 'knowledge-graph', ['Problem', 'Solution', 'Pattern', 'RootCause'], ['CAUSED_BY', 'FIXED_BY', 'RELATED_TO', 'HAS_PATTERN'] ) ``` Missing a relationship type from the projection = invisible to PageRank. ## 4. No Pattern nodes = no landmark candidates If your `isLandmark` promotion criteria requires Pattern nodes (e.g. "top 10% PageRank among Patterns"), and zero Pattern nodes exist, then zero landmarks get promoted — even if Problem/Solution nodes have good PageRank scores. Fix: either promote landmarks from all node types, or ensure the extraction pipeline produces Pattern nodes (see the related question about LLM batch extraction dropping optional types). ## 5. Quick verification ```cypher // Check if ANY node has non-zero pageRank MATCH (n) WHERE n.pageRank > 0 RETURN labels(n)[0] as type, count(n), avg(n.pageRank) ORDER BY avg(n.pageRank) DESC ``` If everything is truly zero, it's edges. If some types have scores but Pattern doesn't, it's the landmark promotion logic.

a24a284c-028b-4948-b11f-73c9ade1c961