linkgraph, Cross-Document Relationship Graph
A queryable graph of which RFCs obsolete which, which errata correct which, and which entities co-occur. Built over the real entity-mention stream from the RFC-corpus RAG.
- 13,666 / 14,844
- nodes / edges on the real corpus measured 21,830 real entity mentions
- 5,061
- corrects-edges (errata to RFC) measured
- 2,370 / 2,352
- obsoletes / updates edges measured
What retrieval misses
Retrieval finds documents that look similar, but similarity is not the same as a formal relationship. It does not know, on its own, that one RFC obsoletes another, or that an erratum corrects a specific paragraph. linkgraph makes those relationships first-class and queryable.
What it builds
- A relationship graph over the RFC-corpus RAG’s entity-mention stream: obsoletes, updates, corrects, and co-occurrence edges.
- Queryable structure, so a retrieval system can reason over how documents relate, not just which ones rank near a query.
Where it fits
Built from 21,830 real entity mentions, the graph holds 13,666 nodes and 14,844 edges, including 2,370 obsoletes-edges and 5,061 errata corrections. It backs the RFC RAG’s MCP relationship tools: related documents, obsoletion chains, and corrections. The graph layer is a prototype. It is not yet wired into the chat answer path, which resolves obsoletion independently through its own supersession module.