Understand source selection
A semantic query can draw on stored graph data, reviewed lakehouse mappings, graph analytics, ontology facts, or vector search. The unified_query_fabric result metadata describes which source and execution strategy were selected.
Read the selection
| Value | Meaning |
|---|---|
|
Stored graph facts. |
|
An eligible reviewed mapping over external data. |
|
A graph algorithm such as PageRank or a weighted path search. |
|
A specialized bounded graph operation. |
|
Ontology-backed reasoning. |
|
Vector-index retrieval. |
|
An available general execution path used when a specialized path is unsuitable. |
A fallback is an execution choice, not automatically an error or an empty result. Check its reason, the returned rows, and the source identifiers before interpreting the answer.
Check a lakehouse contribution
Look for the selected mapping IDs, source hashes, requested fields, row counts, and source-read diagnostics. A semantic mapping must cover the required concepts and constraints; an aggregate also needs appropriate measure and grouping support.
If lakehouse was skipped, inspect the reason before changing the question. A missing mapping or incompatible shape requires a mapping or query change. An empty source-backed result means something different from an unselected source.
The lakehouse tool accepts supported semantic requests and explicit source-bound Cypher/SPARQL projections over published mappings. The regular language endpoints recognize those same source-bound forms. Other Cypher and SPARQL queries use stored graph/RDF data. See the query forms and verification steps before consuming source-backed results in an application.
Reproduce an answer
Record the selected database, profile and alignment hashes, mapping IDs, source versions, query parameters, and result hash. Compare those identifiers first when a repeated question returns different results. Use Results And Inspection for the console view of this metadata.