BINarratorConfidence isn’t a label. It’s a process you can inspect.
Every figure in a BINarrator.ai report starts in a source document and keeps its reporting period. Software re-checks it, eight independent reviewers challenge it, and a named person approves it before it is published. This page shows each step, and what happens when something does not match.
Examples use Meridian Industrial Holdings, our fictional sample company.
Seven checkpoints between a source document and your screen.
Automation does the reading and the arithmetic. People make the decisions that need judgment. If it can’t be shown, it isn’t published.
- 01
Collect
Public sources are checked every day. Each document is stored exactly as published, with a digital fingerprint that proves it was never altered.
- 02
Extract
Figures come from the document’s structured data and ratios are calculated, not guessed. Text is read one section at a time, and every proposed claim carries the exact words behind it.
- 03
Verify
Each quote is searched for in the stored document, word for word. A claim with no match is thrown away; figures are re-checked and the arithmetic re-run.
- 04
Challenge
Eight specialist reviewers re-check the whole draft before a person sees it. A serious finding blocks the report until it is fixed.
- 05
Review
A named reviewer signs in with a passkey, reads each summary beside its quote and the findings, and approves or rejects. Covenants and guarantees always need a person.
- 06
Publish
Approved claims join the report with their quote and link. Anything still missing is listed as a gap, never guessed.
- 07
Watch
A daily check reads new event reports and press releases. A new annual or quarterly report starts a fresh draft that goes through every checkpoint again.
Eight reviewers, eight different ways to catch a mistake.
Each reviewer checks one part of the report against the source documents, apart from the software that wrote the draft, so a mistake is not graded by the process that made it.
Financial figures
Re-calculated from the reported inputs, with periods, units and signs checked.
Debt and liquidity
Every instrument, facility and lease accounted for; the bridge to total debt has to add up.
Covenants and legal terms
Thresholds, compliance statements, guarantees and security read against the agreement’s own words.
Operating measures
Each measure held to the company’s own definition.
Events after the year-end
Financings, deals and leadership changes dated, and marked announced, signed or completed.
Credit conclusions
Every strength and risk must be supported by the figures beside it.
Brief against brief
The executive brief and the credit brief must state the same figures, dates and statuses.
Source completeness
Every report and press release since the annual report must be read, and what they disclose must appear.
What happens when a number doesn’t reconcile?
An illustration with our fictional sample company. The steps are the ones every report follows.
Total debt at December 31, 2025 was $3.9 billion.
- Draft kept $3.9 million, unchanged
- Finding Financial figures reviewer · serious · wrong unit
- Corrected $3.9 billion, re-checked against the source
- Approved by a named reviewer · published with its quote
Publication is blocked
The financial-figures reviewer finds that the draft gives total debt as $3.9 million. The source says $3.9 billion: the draft used the wrong unit. A serious finding blocks the report, so nothing is published.
A correction never overwrites history: the original draft, the finding, the correction and the approval are all kept.
Clear about what software does, and what people decide.
Checks that run the same way every time
- Collects the source documents and fingerprints them
- Takes figures from structured data and re-runs the arithmetic
- Matches every quote to the stored document, word for word, and throws away claims that don’t match
- Compares the executive brief with the credit brief
- Puts the draft through eight specialist reviews before a person sees it
Decisions with a name on them
- A named reviewer approves or rejects every report, signing in with a passkey
- Covenants and guarantees always need a person
- Material changes need a person’s approval
- Nothing is published automatically
- Every approval and edit is recorded, and the record cannot be erased
Every release is tested on real companies before it ships.
- Each release is run over about a hundred real public companies across seventeen sectors.
- Hundreds of figures with known answers are compared. A release that newly gets any of them wrong does not ship.
- A separate set of companies is held back and used only to measure quality on companies the checks were not tuned on.
The seventeen sectors every release is tested on
- Restaurants
- Airlines
- Hotels
- Banks and cards
- Insurers
- REITs
- Utilities
- Energy
- Software and tech
- Retail
- Industrials
- Telecom and media
- Healthcare
- Consumer
- Autos
- Materials and chemicals
- Distribution
Trust is stronger when the boundaries are visible.
- A figure the documents do not give is shown as a gap. It is never estimated.
- Annual figures stay as reported. Later developments keep their own date and status: announced, signed or completed.
- Reports cover U.S.-listed public companies, and the rivals, partners and suppliers named in their own documents.
- Reports support your credit analysis. They are not investment advice and do not replace your own judgment.
What you can check for yourself
Bring your questions. Inspect the evidence.
Bring a company you follow to a walkthrough, or open the sample report and check any figure yourself.