Guide
What the lab is, and how to read it.
Short articles on the ideas behind the lab and the things its pages show. Each starts with the gist, in a nutshell, then goes deeper. A beside a term anywhere on the site opens its article.
Start here
About the earnings lab
The earnings lab reads the quarterly earnings releases US companies file with the SEC, turns them into a knowledge graph of facts that each cite their source, and lets you explore the graph or ask questions of it. It is a research project into trustworthy AI over documents, not an investment tool.
How to read the graph
Each coloured cluster is a sector of companies, and the lines between them say how companies relate. Click anything, a company, a sector or a line, to see what it is and where the evidence came from.
An architect's view
For a technical audience: enterprise and solution architects. The lab is a working reference for a governed semantic layer for AI: an ontology as the single contract, verified extraction with provenance, agents acting as the user through MCP, evaluation as a component, and cost, licensing and abuse controls built in.
The earnings graph
Companies outside the run
The grey dots outside the ring are listed companies the lab has no facts for, drawn only because a company in the run is linked to them as a supplier or customer. They sit beyond their sector, as estimated from their SIC code.
Dependency strength
Amber goes from dim to bright, and lines from thin to wide, as a line stands for more company links. In the season replay, the brightest amber marks a read-through that is due instead.
Earnings read-through
When one company in a supplier link reports its results before the other, its results may say something about the other's, for example a big customer's sales hinting at its suppliers' orders. The lab is testing whether that holds; it is a hypothesis, not a signal to trade on.
Link types
A line's colour says what kind of relationship it is. Amber means a supplier dependency and nothing else; the quieter colours show structure. Lines have no arrows, and clicking one says which way it runs.
Price charts and the market view
With the Pro plan, resting on a company shows its price history in the card, and the Market view shows every company's history on one page, as thumbnails or as one sector of live charts.
Sector roll-ups
Zoomed out, supplier links are added up by sector, one line for each pair of sectors and a ring on a hub for links within a sector. Select a sector to keep its lines while you zoom in.
Sectors and the ring
Companies are grouped by GICS sector, and the sectors sit around a ring in alphabetical order, each a hub with its companies around it. A sector keeps its colour and its place whatever else changes.
SIC and GICS
GICS is the sector classification the run uses for its own companies. For companies outside the run, the page estimates a GICS sector from the SIC code the SEC assigns, which is rough at the edges.
Supplier links
A supplier link means a company named another as a major customer in its annual report (10-K). It is a lower bound, because many companies describe big customers without naming them.
The chat
The chat answers questions about the companies in the graph. It turns your wording into the ontology's terms, queries the graph and the filings, and cites what it used, which lights up in the graph. It needs an access token, can be wrong, and is not investment advice.
The reporting season
The season replay steps through a quarter day by day, lighting each company on the day its results reached the market and counting the read-throughs due. It knows release dates in hindsight, so it shows the past, not what is coming.
The ontology
The ontology
The ontology is the lab's vocabulary for earnings: the kinds of thing a release can state, such as a reported value, guidance or a risk, and how they connect. Every node on the graph is typed by one of its classes, which is why a node's panel names its class.
Analyst questions
An analyst question is one an analyst put to management on an earnings call, with its topic, the concern behind it, its tone and how directly management answered. They show what the market is focused on, which a release never says. Transcripts are licensed for private research, so these appear only in the private view.
Corporate events
A corporate event is a discrete action a company announces alongside its results: an acquisition, a divestiture, a share repurchase, a dividend, a restructuring or a change of leadership. It records who announced it and, for a deal, the other company involved.
Drivers and risks
A driver is a reason management gives for a change in a metric, such as higher prices lifting revenue. A risk, or headwind, is something it says could hold a metric back. Both name the metric they affect, whether the effect is positive or negative, and the part of the business they concern.
Extraction runs
An extraction run is one pass of the lab over a set of filings, turning their text into facts. Every fact records the run that produced it, and the run records the model and prompt version used, so a result can be traced and compared with a later run.
Guidance
Guidance is a company's own forecast for a metric in a future period, given as a range or a single figure, together with what it did to its earlier forecast: initiated, raised, maintained, lowered or withdrawn. It is what the company expects, not what will happen.
Management quotes
A quote is something an executive said, recorded word for word from the filing, with who said it. The lab checks that each quote appears verbatim in its passage, so the words are the company's, not a paraphrase.
Passages
A passage is a short, continuous piece of a filing, such as a paragraph of an earnings release. Every extracted fact links to the passage it came from, so you can read the company's own words behind any number on the graph.
Reported value
A reported value is one number a company stated in its results, such as quarterly revenue or diluted earnings per share, tied to the metric it measures, the period it covers and whether it is GAAP or adjusted. It is the most common fact in the graph.