CompanyGraph
A number does not become a claim in one jump.
How CompanyGraph turns reported numbers into bounded observations, evidence, trust, and reader-facing claims.
In CompanyGraph, reported data crosses a row of stages before a reader meets it — and what readers do with the result flows back and changes the earlier stages:
Each stage answers a different question. None is allowed to borrow certainty from the others.
This separation matters because financial pages can easily make a narrow observation sound like a broad conclusion. CompanyGraph keeps the stages apart so the claim cannot grow stronger simply because the interface sounds confident. This page walks the path a number travels.
Data is not meaning
The easy path is to jump straight from data to confident prose: a figure appears, and a sentence tells you what it means, with no visible line between the number, the check, and the interpretation. CompanyGraph tries to refuse that jump. Raw data, a specific claim, its proof, how far the claim can be trusted, and where a reader may meet it are kept as separate stages — so a weakness at one stage cannot hide behind strength at another.
What each stage does
Each stage answers one question and prevents one kind of false confidence. The question is written under each stage.
01 · Data
What the company reported: raw statements, figures, periods, and source dates.
What did the company actually report?
02 · Quality
Whether a figure is usable at all: units, currency, plausibility, missing data, and freshness are checked before anything is computed from it.
Can this number be used safely?
03 · Claim
A specific observation generated from the data: one bounded pattern, described plainly. Never a verdict, a forecast, or a recommendation.
What does this figure show, exactly?
04 · Proof
The observation is recomputed independently from stored inputs, to see whether it reproduces.
If CompanyGraph checks it again, does it hold?
05 · Evidence
The check becomes a durable record: source, inputs, arithmetic, date, and result.
What is the standing record of support?
06 · Trust
How firmly the system may stand behind the observation — a level that reflects which checks passed, not whether the company is good or bad.
How firmly may the next stage speak?
07 · Placement
Where, if anywhere, the claim may appear: a headline, supporting evidence, a footnote, a stock-page note — or not shown at all.
What reader-facing role is this claim allowed to have?
08 · Surface
What actually appears: a page, a report, a receipt, a caption — with wording and visual register matched to the claim's support.
What does the reader see, and in what register?
09 · Use
A person reads it, compares it, questions it, or takes it as material for judgment.
What happens when the claim meets a reader?
10 · Feedback
What comes back: confusion, a stale figure, a wrong label, a contradiction, wording that sounded stronger than intended.
What did the system learn from being used?
11 · Steering
The owning stage changes: a better data guard, a corrected formula, a clearer caption, a lower placement, a different trust level, a new check.
What changes so the same class of mistake is harder to repeat?
A loop, not a pipeline
The chain does not end at the page. Use creates feedback. Feedback drives steering. Steering changes the earlier stages.
Many pipelines stop at output. CompanyGraph tries to treat output as contact with reality: a reader may notice confusion, stale data, wrong wording, missing context, or a claim that sounded stronger than intended. That feedback is routed back to the stage that owns the problem — the data guard, the formula, the caption, the placement — so the next pass through the loop behaves differently. Nothing is frozen; the system moves toward being well-calibrated, not toward saying more.
Checked does not always mean shown
Proof and placement are different questions. An observation can be reproducible — recomputed cleanly from stored inputs — and still not be suitable as a reader-facing headline. It may become supporting evidence only, or not appear as a claim at all, if its wording would imply more than the evidence supports. Trust is earned by checking; being shown is earned separately, by relevance and plain, accurate wording.
To understand why CompanyGraph keeps these boundaries — and the failure modes they exist to resist — read How CompanyGraph handles claims.
What the financial lens can and cannot see
This whole path begins from reported financial data, and that starting point is itself a boundary. Financial statements are traces of a company, not the company itself. They can reveal useful operating traces — scale, margins, cash generation, leverage, capital intensity, cyclicality, deterioration, consistency, and change over time — but a company is also an operating role in the economy, with dependencies, customers, infrastructure, inputs, and exposures the statements only partly show.
CompanyGraph treats companies as real operating systems, not just financial instruments — and tries to say plainly, at each stage, what the financial lens can support and what it cannot.
Displayed does not mean decisive
When CompanyGraph displays an observation, it is not saying the observation is decisive for every reader, every decision, or every context. Displayed means: this may be useful material to inspect. It does not mean: this is the most important fact about the company.
A historical observation may be accurate and still not be the most important thing about the company today. CompanyGraph shows observations as material for judgment, not as judgment itself.
Output with boundaries
No stage claims truth. Each stage only says how firmly the next one may speak.
CompanyGraph does not ask readers to trust a single confident voice. It shows the path from reported data to bounded claim, and keeps the boundaries visible where it can.
The goal is not more output. The goal is output with boundaries.