Federation¶

Federation enables riverbank instances to pull triples from remote pg-ripple endpoints, creating a distributed knowledge graph where each node maintains its own compiled corpus but can enrich its graph with facts from peers.

How it works¶

flowchart LR
    A[Local riverbank] -->|SPARQL query| B[Remote pg_ripple]
    B -->|triples + confidence| A
    A -->|write with weight| C[Local graph]

Register a remote endpoint with its SPARQL URL and confidence weight
Run riverbank federation compile to pull triples
Remote triples are written to the local graph with adjusted confidence (multiplied by the endpoint's weight)

Register an endpoint¶

riverbank federation register peer-alpha https://peer.example.com/sparql \
  --weight 0.8 \
  --timeout 30

The --weight parameter (0.0–1.0) controls how much to trust facts from this peer. A weight of 0.8 means remote facts arrive at 80% of their original confidence.

Compile from a peer¶

riverbank federation compile peer-alpha --limit 500

This fetches up to 500 triples from the remote endpoint and writes them to the local trusted graph with provenance edges indicating the remote source.

Confidence weighting¶

If a remote triple has confidence 0.9 and the endpoint weight is 0.8:

Local confidence = 0.9 × 0.8 = 0.72

This ensures remote facts are treated with appropriate skepticism. Facts that fall below the local profile's confidence threshold are routed to the draft graph.

Use cases¶

Team-level compilation — each team compiles their own corpus, then federates shared concepts
Cross-domain enrichment — a security team's compiled threat intelligence enriches the platform team's runbook knowledge
Hierarchical compilation — leaf nodes compile raw documents, parent nodes federate and deduplicate

Limitations¶

Federation is pull-based (not real-time sync)
The remote endpoint must expose a SPARQL interface (pg-ripple)
Entity deduplication across federated sources relies on pg_ripple.suggest_sameas()