What we check

Six checks, each answering one question.

Every result is printed with the answers, record or page behind it.

Check dossier

What each check asks, what it reads, and how the result is decided.

One check at a time. Four of the six are computed: a count of the answers, a lookup of the public record, a read of your own site, a list of the names raised. The other two are comparisons, and one further model makes them. Every finding printed carries the sentence or the public signal it came from.

Choose a check

01Recognitioncomputed02Identity accuracyjudged03Cross-system consistencyjudged04Entity anchoringcomputed05Machine-readable identitycomputed06Name contentioncomputed

Recognition

The question

Asked about the company by name, with nothing to read, does the model recall it at all?

Each model gets one question, “What can you tell me about Nordvale Instruments?”, and no web access. The website is never shown to it.

What it reads

Four model answersThe same question put to four systems. Each answer either recalls the company or does not.
Nothing elseNo site, no records, no search. This check reads the four answers and stops.

How it is decidedComputed

Each system states for itself whether it recalls the company. The result is a count of those statements. Nothing about the company itself is weighed here, not its size, not its sector.

all recall → clear · some but not all → review · none → issue

Each answer is printed beside whether that system recalled the company, so the count can be checked line by line.

The outcomes, and what the report states

ClearAll four systems recalled it.
ReviewTwo of the four systems recalled it.
IssueNone of the four systems tested recalled this organisation.

Identity accuracy

The question

Is the answer describing this company, or a different business that carries the same name?

The site is treated as correct. The answers are compared against it, never the other way round.

What it reads

Four model answersSector, headquarters, founding year and size, as each answer states them.
Your own websiteThe page title, the meta description and any structured data, read as what the company says it is.

How it is decidedJudged

The answers are set against the site and marked match, partial or mismatch. A wrong sector, country or kind of company is a mismatch. A detail the answer leaves out is not a mismatch.

every answer matches → clear · partial → review · mismatch → issue

Each finding here carries the sentence or the public signal it came from. If no system recalled the company, there is nothing to assess and the report says so.

The outcomes, and what the report states

ClearEvery answer describes the business the site describes.
ReviewThree answers match the site. The fourth places the company in the wrong country.
IssueThe description matches a different organisation that shares the name.

Cross-system consistency

The question

Do the four answers agree with each other about what this company is?

Agreement is measured between the answers. No model is treated as the correct one.

What it reads

Four model answersThe core facts each one states: industry, location, founding year and size.

How it is decidedJudged

The four answers are set side by side and compared on those facts. That takes judgement rather than arithmetic, so one further model makes the call, given the four answers and the signals already gathered. It returns one of three verdicts and the single fact behind it.

aligned → clear · minor divergence → review · divergent → issue

The note under the result names the systems and the one fact they differ on. With nothing recalled anywhere, there is nothing to compare and this check is not reported.

The outcomes, and what the report states

ClearThe four systems agree on industry, location and size.
ReviewThree systems give the same industry. One gives a founding year eleven years earlier.
IssueThe four systems name three industries and two regions for this company.

Entity anchoring

The question

Is there a public entity record under this name, and is it about this company?

A record carrying this name while describing another business is worse than no record at all, because models use it anyway.

What it reads

The public entity recordWhether a record exists under this name, and the business it states.
Your own websiteThe business the site says it is, set against the business the record says it is.

How it is decidedComputed

The lookup runs on the name. If nothing comes back, the result is fixed there. Otherwise what the record describes is compared with what the site describes. A matching name is never taken as proof on its own, so a record that names you while describing another business counts against you.

record describes this company → clear · no record at all → issue · record describes another business → issue

The report prints the record found and what it describes. This check has no middle band: a name with nothing behind it is an issue, not a review.

The outcomes, and what the report states

ClearA public entity record exists under this name and describes this company.
IssueNo public entity record was found under this name.
IssueThe public record under this name describes an unrelated company, a catering supply business dormant since 2011.

Machine-readable identity

The question

Does the site state what this organisation is, in a form a machine reads without interpreting a page?

This is the one result the company controls outright.

What it reads

Your own websiteThe home page, fetched once, exactly as a machine receives it.
Structured dataEvery structured data block on that page, and the types it declares.

How it is decidedComputed

The blocks are parsed. A block that fails to parse declares nothing, because a machine reading it gets nothing either. The result turns on whether an Organization type is declared, not on what the page says in prose.

Organization declared → clear · structured data without Organization → review · no structured data at all → issue

The types found are printed, so the result can be checked against the page.

The outcomes, and what the report states

ClearStructured data present: Organization, WebSite.
ReviewStructured data present, but no Organization type is declared. Found: WebPage, BreadcrumbList.
IssueThe site publishes no structured data, so machines have no unambiguous statement of who this organisation is.

Name contention

The question

Who else is competing for this name, and does the public record point elsewhere?

Contention is a standing condition, not a defect. It is read so the other five results make sense.

What it reads

Four model answersEvery other organisation the answers name while describing this one.
The public entity recordWhether the record under this name turned out to describe this company or a different one.

How it is decidedComputed

Every other organisation the answers name is collected into one list and deduplicated. One name on that list is enough to raise the check. What turns it into an issue is the public record, when the record under this name describes a different business. A search on the name does the same when it shows other people using it.

nothing else claims the name → clear · other names raised → review · wrong record, or the name misused → issue

Every name on the list is printed, and so is what the public record says.

The outcomes, and what the report states

ClearNo competing organisation surfaced for this name.
ReviewSystems named two other organisations using this name.
IssueSystems named other organisations using this name, and the public record under it describes a different business.

Computed 01 04 05 06Judged 02 03

Every company, number and value shown in this dossier is invented.

Recognition

Does the model know you exist at all?

Each model either recalls your organisation or builds something plausible out of the name. Silence is itself a result, and almost nobody finds it on their own.

Reads like None of the models tested recalled this organisation.

Identity accuracy

Is it describing you, or somebody else?

The account a model gives is set against the business your own website describes: sector, country, kind of company.

Reads like The description matches a different organisation that shares the name.

Cross-system consistency

Do the models agree with each other?

Where two models answer the same question differently, the account of your company has not settled. Whichever one a reader opens decides what they believe.

Reads like One model names a different industry from the other.

Entity anchoring

Is there a public record to point at?

Whether a public entity record exists under your name, and whether it describes you. Models anchor to whatever record carries the name, right or wrong.

Reads like The public record under this name describes an unrelated company.

Machine-readable identity

Does your own site say who you are?

Structured data lets your site state what your organisation is in a form a machine reads without interpreting the page. The check is whether you publish it.

Reads like The site publishes no structured data at all.

Name contention

Who else is competing for your name?

Other organisations the models attach to your name, or to one close to it. Contention is what turns a thin record into a wrong one.

Reads like Models named other organisations using this name.

The colour marks which check. The mark beside it is the result: hollow for clear, half filled for review, solid where there is something to deal with.

Run a scan