About the earnings lab

What it does

Every quarter, listed US companies publish an earnings release: their results, their outlook, the reasons they give, and comments from management. The lab collects those releases from the SEC, and an AI model reads them against a fixed vocabulary, the ontology. What comes out is a graph of facts (reported figures, guidance, drivers, risks, corporate events, quotes), each tied to the company, the period and the exact passage it came from.

This page draws that graph. You can open a company to see its facts, follow the links between companies, replay a reporting season day by day, and ask the chat questions about what the companies said.

Why it exists

AI can read documents quickly, but an answer is only useful if you can tell where it came from and whether it is right. The lab is a working test of how to get there:

• Every fact cites its source. Nothing enters the graph without the passage it was read from.
• Answers are checked. Numbers are checked against the passage, and reported GAAP figures against the structured data companies file with the SEC.
• Wording resolves to meaning. "Sales", "net sales" and "top line" all mean revenue, so a question finds the facts however a company phrased them.

SEC filings make a good proving ground because they are public, consistent in form, and come with a ground truth to score against. The same approach applies to contracts, policies, reports, or any large body of documents an organisation needs to reason over.

What it is not

The lab is research. Its facts are extracted by AI and can be wrong, and nothing on this page is investment advice. Where it explores ideas about markets, such as the earnings read-through, they are hypotheses under test.

Where to start

Read how to read the graph, then click a company. For how the lab is built, and the AI and architecture ideas it explores, see an architect's view.