Knowledge graph

Show measured co-occurrenceReset view

The document, as a map

Arrows are claims the document makes. Circle size is how much of the document is spent on that subject.

Select anything to see what it says and where.

Browse as a list instead

What you are looking at

The map is built from the document rather than alongside it. Every subject on it was found in the text, its size is how many passages discuss it, and every link leads back to the passage it came from.

There are two kinds of connection, and the difference matters.

Solid arrows are claims. They are relationships the document actually argues — that galcanezumab failed in chronic cluster headache, that the manufacturer’s collapse stranded SPG-stimulation patients, that Germany has the strongest oxygen position reviewed. Each one carries a one-line statement of the claim and links to the passages behind it. There are around fifty, and together they are the spine of the document’s argument.

Dotted lines are measured co-occurrence. They mean the document repeatedly discusses two subjects in the same passage, filtered so that only associations stronger than chance appear. That is genuinely informative — it is how you notice that Denmark shows up wherever registry data does — but it is a pattern in the text, not an assertion about the world. They are off by default.

Nothing on this map was generated by a language model. The vocabulary is a curated list, the measurements come from the text, and the claims were written by hand against the sections they point at. That is deliberate: a machine-inferred relationship rendered as a confident arrow is exactly the kind of thing that reads as fact and is not.

Ways to use it

Search for a treatment you are considering and read its neighbourhood — what it targets, which trials tested it, which form of the disease it works in, and where you can actually get it.

Select Chronic cluster headache and look at how many arrows arrive labelled failed in, dose-limited in or last resort in. That asymmetry is the single most important pattern in the document, and it is more legible here than in prose.

Turn on co-occurrence and filter to jurisdictions to see which countries the evidence base actually comes from — and, more tellingly, which ones it does not.