Context Graph Source Authority Ranking

Not every source that can answer a question deserves the same authority.

Authority is a graph problem

A context graph connects chunks, documents, tables, owners, lineage, policies, quality signals, and answers. Source authority ranking adds a practical question to that graph: which source should an agent trust first for this kind of answer?

The system of record may own one fact. A derived table may own another. A policy document may govern access. A vector index may only be a retrieval copy. Ranking authority means distinguishing those roles before answer generation.

Ranking needs governance signals

Authority should not be a popularity contest between chunks. It should include source type, owner, lineage distance, freshness, policy boundary, quality score, consent state, and incident history.

That is why lineage and provenance standards matter. They help connect the retrieved artifact to the system that produced it and the owner who can defend it.

The ODI pattern keeps authority outside one index

Open Data Infrastructure makes authority portable by keeping metadata, lineage, policy, and ownership available across tools. A vector index can use those signals, but it should not become the only place authority is defined.

For adjacent reading, see context graphs for retrieval governance, context graphs for AI root cause analysis, and AI-ready context lineage fingerprints. Authority ranking is the bridge between retrieval quality and governance.

What breaks first

Authority failures look like good retrieval with bad judgment.

• A stale document outranks a current system-of-record table because the text matches better.
• A derived table is treated as authoritative without lineage to its source.
• The owner field exists in a catalog but never reaches retrieval ranking.
• Policy documents and operational data disagree, and the agent has no rule for conflict.

Questions to ask before ranking

Ask which source is authoritative for each entity, metric, policy, and time range. Ask how the graph handles derived sources, copied indexes, stale documents, and source conflicts.

A context graph earns trust when it can explain why one source outranked another.

Sources to start with

These primary sources anchor the technical claims in this guide.

• W3C PROV overview
• OpenLineage object model documentation
• DataHub lineage documentation
• Open Policy Agent documentation

A context graph should not only find evidence. It should know which evidence has authority.