Aduhelm
Accelerated approval for Alzheimer's, over the advisory committee's objection.
Drug approvals. Plane crashes. Autonomous systems. Rebuilt from the primary record, so you can see how the story actually happened.
The page is a stand-in. The highlighted entities are real, taken from the built graph.
The Ephemeris
This one is 392 pages: the FDA's medical review of Aduhelm, the Alzheimer's drug approved in 2021 over its advisory committee's objection.
Read once
Every drug, trial, dose, endpoint and adverse event becomes a node, tied to the document and page it came from.
One drug
Counted as they are drawn. Every dot on screen is a real entity from the Aduhelm graph.
The queue
Leqembi, Kisunla, the advisory committee record and 22 more Alzheimer's drugs are lined up. 5,015 pages wait in the queue, read at a measured 16.47 pages a second.
Then the next record
283 companies after a Complete Response Letter: 456 letters and 707 SEC filings, read into the same graph. The same method, a second institution.
Being read now
The institutions
Close to home
The medicine prescribed to you. The plane you boarded. The systems making decisions around you. Behind each is a public record of how it came to be.
The cases
The map shows the whole record. A reconstruction tells one story from it, event by event, every claim traced to its source.
Accelerated approval for Alzheimer's, over the advisory committee's objection.

Two FDA letters, then accelerated approval: 16 questions answered from the FDA's letters and the company's SEC filings, every quote traced to its source.
The method
Primary documents only: reviews, letters, dockets, filings. Each keeps its citation, retrieval date and content hash.
Every document is read once into a knowledge graph of entities and the relations between them.
Every claim points back to the document and page it rests on. Nothing summarized away.