Visualising Semantic Graphs: Avoid The Blob, Use Verb Edges, And Flip The Subgraph
Everything the graph explorer does, and three things it still does not.
team/humans/dinis_cruz/briefs/06/10/network-intelligence/v0.33.16__arch-brief__sg-send-semantic-graph-visualisation-subgraph-flip-verb-edges.md
Commitb561fa6b14 · carried 23 August 2026
LicenceApache-2.0 (repository) · CC BY 4.0 stated in the file
Bytes19,755 · 3,143 words · SHA-256 3e491bf9c8564f00
Dated10 June 2026
Why it is here
A graph visualisation that renders everything renders nothing: it collapses into a hairball with no readable structure. This document is the method for avoiding that, and it is the closest thing the corpus has to a specification for the explorer on this site: rich nodes are good, edges are two-way verbs and never relates-to, build the universe in a first pass, then flip the query around a found node to get a small, relevant graph.
What it says
- The blob anti-pattern: rendering the whole graph produces something with no readable structure.
- Rich nodes are good; the answer is not fewer properties, it is fewer nodes on screen.
- Two-way verb edges, and never a generic association edge.
- The subgraph flip: find a node, then re-query around it for a small, condensed graph.
- The LLM builds and prunes the query; the final query stays pure node-and-edge mathematics.
Worth re-reading for. The subgraph flip is the one big idea the explorer has only partly built. Radius exploration is the flip with a fixed depth rather than a re-query.
Where this document lands on the site
The document is the peak. Below it hang the concepts measured in its text and the places on this site that rest on it — and the two are different kinds of link, drawn differently: a measured edge is dashed and carries its count, an authored edge is solid. The method is stated in full on the hub.