Fly through the graph
The knowledge graph is usually something you search one entry at a time. Here it is all at once: every claim as a point, every citation, contradiction and part-of relation as a line. Drag to move, scroll to zoom, click a point to read what it is and what it rests on, and use the filters in the top bar to keep only the types you want.
This view is drawn by Graphistry, which runs on their servers and loads in a frame.
Nothing loads from them until you ask it to.
What you are looking at
Point size is how many relations a thing has, so the large points are the load-bearing ones: a review cited by a hundred claims, a structure half the literature is about. Colour is the kind of thing. The lines are the relations — a claim citing its source, a claim contradicting another, a nerve and what it supplies, a gene and where it is expressed.
Two honest notes about the picture. First, a force layout puts related things near each other, but distance on screen is not a measurement: do not read anything into how far apart two points are. Second, what you are seeing is our reading — how much of the literature we have entered, not how much exists. A sparse corner means we have read less there, not that less is known.
How it is built
The graph lives in this project as two plain text files, one JSON object per line, and you can download both. A script turns a chosen slice of it into a table of points and a table of lines, and uploads that to Graphistry, which draws it on a GPU. The slices are made by filtering node types and relations — the nervous-system view, for instance, keeps neural structures, the nerves and what they supply, the neural genes and where they are expressed, and the claims and questions about all of them.
One decision is worth naming because it changes the picture completely. Species membership used to be a relation, and 2,477 of the graph's edges pointed at just 92 species nodes; in a force layout those became the entire image, dragging every claim about Hypsibius exemplaris into one knot. It is now a property of a node instead, so you can still filter by it but it exerts no pull. The general rule, for anyone drawing their own graph: a relation that is really a category should be an attribute.
The code is tools/graphistry_view.py, and the views, their sizes and the reasoning are written up in the project's docs/GRAPHISTRY.md. Everything except the drawing runs locally; the upload is public, which is why you can see it.
Credits
Drawn with Graphistry, whose author suggested embedding it here — what their tool showed us that our queries did not. The graph itself is CC BY 4.0 for our work; the facts belong to the sources cited on every line.