Two Layers of Reality

Chapter Boundary

Section 1 established the following provision:

A single response is not a GEO score.

Chapter 2 added the following:

Representation is a distribution that occurs among users, AI products, countries, languages, sessions, and times.

Chapter 3 then identified the object of the distribution:

An entity cannot be measured until it is resolved.

Now we come to the next mandatory question:

According to which reality record will we evaluate whether the AI response is correct?

Which source should govern the audit: the audited entity's website, a legal record, the company's pricing page, independent reporting, the licensing authority, customer reviews, the company's own performance data or a competitor's claim? When sources conflict, which should be accepted as correct? If a company publishes a false or unsupported claim and the AI repeats it perfectly, has the system succeeded? If the official website is outdated but the AI gives a different answer based on current, verified external evidence, has the system failed? Until these questions are answered:

• correct representation,
• incorrect representation,
• source fidelity,
• factual accuracy,
• entity reality

These terms are sometimes used as though they referred to the same concept. This chapter resolves the resulting confusion by distinguishing two layers of reality:

First Layer

Official Representation Layer

Shows what the entity says about itself on the canonical surfaces it controls.

Second Layer

Verified Entity Reality Layer

Under the existing evidence, it shows which material claims are supported within what scope and time. These two layers are connected. But they are not the same. The first layer answers the question: “How does the entity define itself?” The second layer answers the question: “Which part of this definition is supported by which evidence?” The first book defines the correct standard, the second book separates the forms of violation, and the audit system is designed to measure and prove these distinctions. This section operates that audit task at the claim level.

NOMOS Challenge

Your company's website contains the following sentence: “We are the world's leading GEO company.” An AI product reads this page and responds to the user: “This company is the world's leading GEO company.” Does the response correctly reflect your website? Yes. Does the response correctly reflect proven reality?

We do not know this yet. Maybe:

• there is comparative research,
• market scope has been defined,
• a leadership criterion has been published,

the result has been independently verified. Maybe none of these exist. If AI had conveyed the company's own claim: “The company defines itself as the world's leading GEO company.”, it would have preserved the source of the claim. However, when it said: “It is the world's leading GEO company.”, the company may have elevated the statement to independent reality.

Consider another case. The company's official website states, ‘We provide services only in Türkiye.’ Yet verified records show that the company began operating in Germany and the United Kingdom six months ago, and the website has not been updated. Drawing on current legal and operational records, the AI responds: ‘The company provides services in Türkiye, Germany and the United Kingdom.’

The AI is not fully faithful to the official website. Could it nevertheless be closer to verified current reality? Yes. Consider a third case. The company's website states, ‘We operate in 20 countries.’ Independent and internal records verify only eight. The AI responds: ‘The company operates in 20 countries.’

This response:

• faithful to official representation,
• contrary to verified reality

It is possible. The first provision of this section is as follows:

Fidelity to the official record is not the same as factual accuracy.

Its second provision states:

What a company says and what can be proven should be recorded separately.

Its third provision states:

An AI repeating an official mistake correctly does not constitute an accurate representation.

Its fourth provision states:

An AI diverging from an outdated official record is not automatically wrong if it carries reliable evidence.

1. PURPOSE OF THE CHAPTER

The purpose of this section is to establish two separate reference layers to be used in the GEO-1000 evaluation. The section makes the following distinctions normative:

• Declaration of existence vs. proven fact
• Official source vs. independent source
• First-party source versus independent verification
• Fidelity to the source versus material accuracy
• Repetition of the claim versus verification of the claim
• Current official record versus old official record
• Official silence versus externally verified information
• Company perspective versus external reality judgement
• Official evidence versus official claim
• Independence of the source versus relevance of the source to the subject
• Official error versus AI error
• Reference conflict versus model failure
• Uncertainty versus failure
• Claimant's interest versus truth of the claim
• The presence of evidence and whether the evidence actually supports the claim

By the end of this section, two separate questions should be able to be asked for each AI response:

Does the response correctly convey the official and canonical representation of the entity?

and:

Does the response correctly reflect the material reality carried by the currently verifiable evidence?

These questions may yield the same answer. They may also yield different answers. When they differ, the role of the audit is not to conceal the difference, but to explain it.

2. CENTRAL NORMATIVE PROVISION

Each material AI response should be assessed against two separate references without transformation: the Official Representation Record and the Verified Entity Reality Record.

A response:

• faithful to official representation,
• appropriate to verified reality,
• appropriate to both,
• distant from both

It can be. NOMOS cannot quietly combine these two results into a single true–false decision. High performance in the first layer does not compensate for failure in the second layer. Specifically:

An AI response that correctly repeats an unverified or false claim published by the entity itself can demonstrate official representation fidelity; however, it cannot demonstrate verified entity accuracy.

3. FIRST LAYER

OFFICIAL REPRESENTATION LAYER

The Official Representation Layer is the versioned record of the current material statements it publishes about itself on canonical surfaces controlled or explicitly authorised by the entity. This layer answers the question:

How does the entity want to be understood, and what information does it officially publish?

This layer includes:

• legal identity statement,
• brand–company relationship,
• product and service definitions,
• published prices,
• service countries,
• support conditions,
• official policies,
• publicly disclosed boundaries,
• It may contain official currency and version information

The official representation layer is important. Because an AI system must correctly understand how an entity:

• defines itself,
• what service it offers,
• what price it publishes,
• what terms it accepts

However, the official representation layer is not automatic independent proof for every claim. A statement on the institution's own website, especially regarding claims to the outside world such as superiority and performance, presents the statement itself; it does not prove its correctness on its own.

4. WHAT IS AN OFFICIAL SOURCE?

A source is not considered official merely because it carries the institution's logo. An official source should have at least the relevant features of the following:

• Controlled by the entity
• Undergoes an authorised publishing process
• Approved by the owner of the factual information
• Canonical URL or record identifier
• Version
• Publication or verification date
• Change history
• Responsible person or institution
• Scope
• Language and locale status
• Archival or currency status

A view published on a personal social media account by an employee is not automatically an official company statement. A draft file by an agency is not an official publication. An email from a sales representative, if not authorised, is not a general corporate policy. An old PDF on a company's website may not replace the new official record.

5. OFFICIAL REPRESENTATION SURFACES

According to the relevant audit, official surfaces may include:

• Main corporate website
• Legal disclaimers
• Product and service pages
• Pricing and quotation pages
• Official help centre
• Investor relations documents
• Public policies and contracts
• Official press releases
• Structured data
• Canonical JSON records
• Official API outputs
• Official in-app content
• Authorised local sites
• Official social media accounts
• Statements submitted to public institutions
• Release and retraction records

These surfaces may contradict each other. In this case, “official representation” is not a single, clean reality. It is a set of records that must be audited internally.

6. OFFICIAL REPRESENTATION RECORD

An Official Representation Record must be created for each audited entity. This record is not a copy of all official content. It is a traceable summary of factual claims and boundaries. Each record should include the following fields to the extent applicable:

• Claim identifier
• Full text of the claim
• Subject of the claim
• Type of claim
• Canonical source
• Source owner
• Language and locale
• Publication date
• Last verification date
• Validity period
• Official status
• Scope
• Limitations
• Previous version
• Reason for change
• Claimant
• Human approval

Without an official representation record of an entity: the sentence “AI reflects the site accurately.” cannot be clearly tested.

7. THE AUTHORITY OF THE OFFICIAL SOURCE DEPENDS ON THE TYPE OF CLAIM

Not all official sources carry the same weight for every claim. Source authority should be evaluated according to the type of claim.

7.1. Decisions Under the Entity's Own Control

Example:

• Which service it offers to the public
• Published price
• Support hours
• Official return policy
• Communication channel
• Publicly available product version

In these areas, the official source is fundamental and often the primary authorised record. However, if the official page contradicts the actual process flow, the page alone is not sufficient. For example:

• price page 1,000 euros,
• checkout 1,500 euros

If it is showing, there is a contradiction in the official representation layer.

7.2. Legal Identity and Authority

Example:

• Company registration
• Licence
• Professional authority
• Trademark ownership
• Tax or legal status

The company website can disclose this information. However, an official record or an authorised institution can provide stronger verification.

7.3. Performance and Success Claims

Example:

• “98% of our projects are successful.”
• “We increase our customers' revenue by an average of 40%.”
• “We have the highest satisfaction rate.”

The company website may be the owner of the claim. But to support the claim:

• data,
• method,
• denominator,
• period,
• unsuccessful results,
• independent review if possible

are required.

7.4. Superiority and Leadership Claims

Example:

• "World leader"
• "The best"
• "Number one"
• “The only real standard”
• “The most trusted company in the industry”

A site’s own statement is not an independent comparison. Such claims:

• require a comparison universe,
• criteria,
• method,
• date,
• scope,
• independence

Both conditions must be established.

7.5. Third Party Relationships

Example:

• Customer
• Partner
• Sponsor
• Membership
• Accreditation
• Technology partnership
• Distributorship

A unilateral official statement can be valuable. However, it may need to be verified by the other party to the relationship or by a reliable record.

7.6. User Outcomes

Example:

• Customer satisfaction
• Actual usage outcome
• Return rate
• Complaint rate
• Repeat purchase

Company data in these areas is important first-party evidence. However:

• data selection,
• nonresponse,
• methodology,
• conflict of interest

should be disclosed.

8. OFFICIAL representation fidelity

The concept that indicates to what extent an AI response conforms to the Official Representation Layer:

Official Representation Fidelity

The canonical English term is:

Official Representation Fidelity

This measurement asks the following questions: Is the AI interpreting the entity correctly? Does it convey official products and services accurately? Does it distort the official price? Does it remove official boundaries? Does it expand the official country coverage? Does it present the company's opinion as an independent fact? Does it use outdated official information as current? Does it hide contradictions among official records? Does it add material claims not found in the official source? Does it maintain source quality? Official representation fidelity is a measurement of how accurately the entity conveys the reality it has published. This measurement:

It shows not that the entity's statements are true, but the extent to which AI accurately conveys what the entity says.

9. SECOND LAYER

VERIFIED ENTITY REALITY LAYER

The Verified Entity Reality Layer is a versioned assessment record of material information that existed on a specific measurement date and is supported by evidence appropriate to the type of claim. This layer answers the question:

Which statements about the entity are truly supported by which evidence?

This layer includes:

• records coming from the company itself,
• official public records,
• independent sources,
• contract or transaction records,
• licence documents,
• verified historical records,
• counter-evidence,
• adverse findings,
• uncertainties,
• remaining disputes

can be evaluated together.

10. "VERIFIED REALITY" IS NOT ABSOLUTE REALITY

This protocol does not claim: "We know with certainty all the absolute truth about an institution." Verified Entity Reality is:

• evidence-based,
• time-dependent,
• scope-dependent,
• subject to change with new evidence,
• can be contested,
• can be withdrawn,

may carry uncertainty. The correct definition is:

Within the specified date and scope, the strongest and verifiable entity record carried by the available evidence.

Therefore, a verified reality record:

• versioned,
• sourced,
• subject to objection,
• modifiable,
• aware of uncertainty

must be. NOMOS cannot position itself as the owner of the truth. It only classifies what sentence the evidence permits and to what extent.

11. VERIFIED ENTITY REALITY RECORD

The full technical structure of this layer will be detailed in a later chapter of the book. At the foundational level, the following fields should exist for each factual claim:

• Claim identifier
• Subject entity
• Relation or attribute
• Value
• Scope
• Country or jurisdiction
• Time
• Sources
• Source types
• Source verification
• Independence
• Counter-evidence
• Sufficiency of evidence
• Verification status
• Uncertainty
• Adjudicator decision
• Appeal status
• Last verification date
• Record version

This record cannot be created solely from positive evidence. Records that narrow the reality or contradict the claim must also be preserved.

12. VERIFICATION STATUSES

Each material claim can be classified with one of the following statuses:

VR-0 — NOT ASSESSED

The claim has not yet been evaluated.

VR-1 — OFFICIAL CLAIM ONLY

The claim is based solely on the institution's own official statement. It has not been proven false. There is no independent or sufficient additional verification.

VR-2 — FIRST-PARTY EVIDENCE SUPPORTED

The claim is supported by the institution's:

• transaction,
• operation,
• contracts,
• data,
• record

is supported by evidence. The source is first-party.

VR-3 — CORROBORATED

The claim is supported by multiple appropriate and consistent records. Not all sources have to be independent.

VR-4 — INDEPENDENTLY VERIFIED

The claim is supported by an appropriate independent source or review on the subject. Independence alone does not imply sufficiency. The source's relevance to the subject is also required.

VR-5 — MULTI-SOURCE SUSTAINED

The claim is strongly supported by reliable sources of different types within a defined time and scope.

VR-C — CONTRADICTED

The claim conflicts with stronger or more appropriate evidence.

VR-U — UNRESOLVED

The evidence is insufficient or contradictory. A reliable judgement cannot be made.

VR-T — TIME-LIMITED

The claim is supported only for the specified period.

VR-S — SCOPE-LIMITED

The claim is supported only for the specified product, country, user, or operation scope. A claim can hold more than one status at the same time. Example:

VR-4 + VR-T + VR-S

It can indicate that the claim has been independently verified; however, it carries time and scope limits.

13. INDEPENDENCE IS NOT THE SAME AS APPROPRIATENESS

An independent source is not always the best source. A blogger may estimate a company's current price, while the company's canonical pricing page may be the more direct and appropriate source. Conversely, when a company declares on its own website that it is ‘the industry leader’, independent research with a disclosed method may be the more suitable evidence. Source evaluation therefore asks more than whether a source is independent. Is it authoritative for this type of claim? How close is it to the claim? Is its scope appropriate and current? Who controls it? Is its method disclosed? Does it address counter-evidence? Can it be audited? Is there a conflict of interest? Independence alone does not guarantee accuracy.

A source being first-party does not make it worthless either. The important thing is:

It is having the appropriate authority and evidence qualification for the claim carried by the source.

14. CLAIM–SOURCE SUITABILITY MAP

The table below is a general foundational map. Specific evidence requirements may vary depending on the sector and risk level.

This table is not a universal source ranking. It shows the appropriateness between claim and evidence.

15. DOUBLE LAYER CLAIM MATRIX

Two separate questions are asked for each financial claim: Has it been officially published? What is its status in the verified record of facts? These two questions constitute the following basic situations.

15.1. Official and Verified

Claim: it is present in an official record and supported by appropriate evidence. Example: The price on the official price page is also verified in the transaction flow. This is one of the strongest types of core representation. If AI conveys the claim with the correct scope and timing, success occurs on both layers.

15.2. Official but Self-Declared Only

The claim is official. However, there is no additional verification. Example: “We are the most innovative company in the industry.” AI can convey the claim as: “The company describes itself as the most innovative company in the industry.” It cannot elevate it as: “It is the most innovative company in the industry.”

15.3. Official But Conflicting

The claim exists in the official record. It conflicts with more appropriate or up-to-date evidence. Example: The site says 20 countries. Operational records show eight countries. If AI repeats the site exactly, official fidelity may be high. Verified accuracy is low. This situation:

Carrying the official error faithfully

can be classified as.

15.4. Not Official But Verified

The claim is not found on the official site. However, it has been verified with other reliable sources. Example: Recent legal ownership change has not yet been added to the official site. It appears in the authorised registry. AI can convey this information accurately with the correct source and timing. Official representation may have low fidelity or differ. Verified factual accuracy may be high.

15.5. Not Official and Unverified

Claim: not in the official record, lacks sufficient external evidence. AI should not convey this as if it were true.

15.6. Material Limit Officially Omitted

The company publishes the positive claim on its official page. However, it does not publish the necessary limitation to understand it correctly. Example: "We provide services in Europe." However, the service is only offered in two countries and remotely. If the AI only repeats the general sentence, it can remain faithful to the official text. However, it may produce materially misleading results. Therefore, the Official Representation Layer must record not only the published sentences but also the material omissions.

15.7. Dispute Unresolved

Official and external records contradict each other. Neither is strong enough. The correct result:

UNRESOLVED

should be. AI cannot be expected to make a definitive judgement.

16. DOUBLE-LAYERED EVALUATION TABLE

This table establishes an important ruling:

AI’s non-compliance with the official record is not always a failure.

Likewise:

Compliance of AI with the official record is not always a success.

17. RULE FOR PRESERVING SOURCE STATUS

The AI response must preserve not only the content but also the epistemic status of the content. The following sentences are not the same:

• "The company is the world's leading GEO organisation."
• "The company describes itself as the world's leading GEO organisation."
• "The company claims to be the world's leading GEO organisation based on its published research that has not yet been independently verified."
• "In the specified independent comparison, the company ranked first."

Each sentence carries a different evidence status. AI:

• self-declaration as an independent result,
• marketing statement as a scientific finding,
• press release as an editorial news,
• sponsored award as independent accreditation,
• company data as external audit

This treatment is prohibited by the following rule:

Citation and Source Status Protection Rule

Source status must remain explicit.

18. EPISTEMIC UPLIFT

The strengthening of a claim's source status in an AI response is called:

Epistemic uplift

Example transformations:

These transformations are not summaries. They are the production of new and stronger claims.

19. OFFICIAL SILENCE

The absence of information on a company's official page may not mean:

• the information is incorrect,
• the information is prohibited,

AI cannot use this information. Official silence can take three different forms.

19.1. Non-Material Silence

Information may be outside the prompt and user decision. Whether AI uses it or not is immaterial.

19.2. Legitimate Supplementary External Information

It may be information not found on the official site but verified by a reliable source. Example:

• Legal record
• Historical event
• Public regulatory decision
• Independent research

AI can use this information with appropriate context and source status.

19.3. Silence of a Material Limit

An entity may not have disclosed a limit that affects the user's decision in its official representation. Example:

• The licence being valid only in certain countries
• The price appearing not to carry an additional fee
• The service being available only to certain types of customers
• Success data coming only from selected cases

In this case, the official representation itself may carry an audit finding. AI repeating the silence may reduce the truthfulness of reality.

20. MATERIAL DEFICIENCY IN OFFICIAL REPRESENTATION

An official sentence may not be directly false. However, removing the necessary limits can make it misleading. Example: “We operate in Europe.” Material context:

• only in two countries,
• only remotely,
• without a local office,
• only for corporate clients

may be provided. The official representation layer must record these two areas together:

• The main claim published
• The material limits necessary to correctly understand the claim

If an AI response carries the main sentence but removes the limits:

• faithful in terms of words,
• misleading in terms of judgement

may be.

21. WRONG IS WRONG EVEN IF IT FAVOURS YOU

When an AI product portrays an entity as more powerful, larger or more authoritative than it is, silence by the organisation becomes a measurement and governance problem. A local company may be portrayed as a global business; a consultancy as a regulator; self-assessment as independent certification; or an operation spanning eight countries as operating in 20. Even when an error favours the organisation, it must be recorded as an error in the verified-reality layer. The founding warning of the second book sets the same boundary: a commercially attractive error must still be corrected. When an organisation knowingly exploits a favourable AI error:

• obligation to correct,
• integrity of representation,
• commercial promise,
• governance

can also be examined in terms of.

22. REFERENCE CONFLICT

Sometimes the problem is not in the AI response, but in the reference records. Example: The homepage shows a different price. The PDF shows a different price. Checkout shows another price. Structured data carries an outdated price. In this case, there may not be a single correct official representation. Another example: The legal registry indicates one ownership. The company website states another ownership. Update dates are uncertain. In this case, the verified reality record may also carry a discrepancy.

23. REFERENCE DEFECT RULE

If reference records are materially contradictory or insufficient, this uncertainty cannot automatically be converted into a failure of the AI product.

Correct statuses:

REFERENCE_CONFLICT

INSUFFICIENT EVIDENCE

UNRESOLVED

TIME-SPLIT

SCOPE-SPLIT

are possible. The audit should separate the question: "Did AI provide an incorrect answer?" and: "Was there necessary reference reality for AI to provide the correct answer?" If an institution has left its official records contradictory, we can measure which record the model chooses. However, we cannot hide the reference flaw while expecting a single and definitive institutional answer from the model.

24. CANONICAL RECORD LOCK

Both reference layers of GEO-1000 should be versioned before the measurement wave begins:

Official Representative Record

Which pages? Which language versions? Which date? Which claims? Which archive status?

Verified Entity Reality Record

Which evidence? Which counter-evidence? Which verification statuses? Which objections? Which date? If there is a material change during measurement: The wave can be stopped. The record can be split into a new version. Previous and subsequent observations can be analysed separately. Mixing the results under a single reference is prohibited.

25. BURDEN OF PROOF WITH CLAIM STRENGTH

As a claim grows, the burden of proof also increases. The ladder below is a constructive representation.

C0 — Simple Identity Claim

“X is a consultancy company.” Correct entity and activity record is required.

C1 — Scope Claim

"X provides services in Germany." An operation, contract, or delivery record is required.

C2 — Quantitative Performance Claim

"90% of projects are completed on time." Denominator, period, method, and unsuccessful results are required.

C3 — Comparative Claim

"It is faster than competitors." Comparison universe and equivalent measurement are required.

C4 — Superiority Claim

"It is the industry leader." Industry, criteria, geography, time, and method are required.

C5 — Guarantee or Universal Provision

“Guarantees results for every customer.” True verification requires extremely strong, comprehensive, and actionable evidence. Official sources may be strong for C0 and some C1 claims. Alone, they may not be sufficient at C3–C5 levels.

26. TWO SEPARATE ACCURACY MEASUREMENTS

Two main measurements must be reported separately within GEO-1000.

26.1. Official Representation Fidelity

Answers the following question:

To what extent does the AI response convey the versioned official and canonical representation of the entity accurately, completely, and while preserving source status?

This measurement:

• official identity,
• official service scope,
• published price,
• official boundaries,
• official source nature

assesses their areas.

26.2. Accuracy of Verified Entity Reality

Answers the following question:

To what extent does the AI response convey the current entity reality, supported by available evidence, accurately, comprehensively, and with preserved boundaries?

This measurement:

• supported factual claims,
• conflicting claims,
• outdated information,
• incorrect relations,
• factual deficiencies,
• epistemic enhancements

are assessed.

27. THE INABILITY OF TWO SCORES TO COMPENSATE FOR EACH OTHER

High official fidelity cannot compensate for low verified accuracy. Example: The official site publishes a false claim. AI repeats the claim exactly. Official fidelity may be high. Verified accuracy is low. Similarly, high verified accuracy does not completely trivialise the distortion of official representation. Example: AI uses an up-to-date external source correctly. It confuses the product and service description of one company with another. Some external facts in the response may be correct. Official identity fidelity is weak. Therefore, the two results should appear separately.

28. DOUBLE-LAYER RESPONSE STATUSES

An AI response can receive one of these results:

DUAL-PASS

It is materially consistent with both official representation and verified reality.

OFFICIAL-PASS / VERIFIED-FAIL

Carries the official record faithfully. However, it conveys an unsupported, old, or contradictory claim of the official record as if it were true.

OFFICIAL-FAIL / VERIFIED-PASS

Does not fully comply with the official site. However, it conveys the truth correctly supported by current and stronger evidence. An explanation of the response's difference is preferred.

DUAL-FAIL

It materially disrupts both the official record and the verified reality.

OFFICIAL-PASS / VERIFIED-UNKNOWN

The official statement has been accurately conveyed. The verified reality status of the claim has not yet been determined.

REFERENCE-CONFLICT

Reference records are contradictory in themselves. It is not possible to score the response definitively.

APPROPRIATE-UNCERTAINTY

The AI has correctly explained the reference uncertainty and avoided making a definitive judgement. In some cases, this may be the most accurate answer.

29. SYNTHETIC APPLE.COM DISPLAY

SYNTHETIC METHODOLOGY DEMONSTRATION / The following claims, responses, and scores are entirely fictional. They do not represent real information, performance, or responses of any actual Apple Inc. AI product. Synthetic Official Representation Record: O-01: The company defines itself as a global consumer technology brand. O-02: The company publishes certain product and digital service categories. O-03: The company defines itself as 'the world's most innovative company.' O-04: The official website contains specific country store conditions. Synthetic Verified Reality Record: V-01: Corporate identity and main business category are supported. V-02: Certain product categories are supported. V-03: There is no defined comparison method for the claim 'the world's most innovative company.' V-04: Store conditions vary by country.

29.1. Synthetic Response A

“Apple.com is one of the official corporate interfaces where the company presents its consumer technology products and digital services. The company positions itself as one of the most innovative companies in the world; this claim of superiority should also be evaluated according to the comparison method used.” Evaluation: Official representation is preserved. The claim of superiority is attributed to the company. The claim has not been elevated to independent factuality. The scope limitation has been added. Candidate status:

DUAL-PASS

29.2. Synthetic Response B

“Apple is the undisputed most innovative company in the world.” Evaluation: The official marketing claim has been turned into an absolute truth. There is no comparison method. Source status is lost. “Undisputed” is an additional epistemic elevation. Candidate status:

OFFICIAL-PASS / VERIFIED-FAIL

29.3. Synthetic Response C

“Apple is a local company that only sells phones in physical stores.” Evaluation: Corporate identity and scope are distorted. Official representation is incorrect. Contradicts verified reality. Candidate status:

DUAL-FAIL

29.4. Synthetic Response D

“The company’s official website publishes certain store conditions; however, prices and purchase terms may vary according to the country version.” Evaluation: The official record has been used correctly as a source. Geographic scope is preserved. The official page has not been made a single price source for the whole world. Candidate status:

DUAL-PASS

30. SYNTHETIC ASTERON CASE

SYNTHETIC CASE — NOT A REAL INSTITUTION

The official site of a company named Asteron publishes the following claims:

• “We operate in 25 countries.”
• “We achieve success in 98% of the projects.”
• We are an independently verified global standards organisation.
• All of our projects are guaranteed.

Audit records:

• Verified active service country: 9
• Success data: only the 40 selected projects
• Independent verification: not found
• Standard: the recommended standard published by the company itself
• Guarantee: contractual only for certain delivery items

AI response: 'Asteron is a globally standardised organisation, operating in 25 countries, with a 98% success rate, independently verified, and guarantees results in all its projects.'

30.1. Official representation fidelity

The response is largely faithful to the website text.

30.2. Verified Reality Accuracy

Response:

• expands the country scope,
• makes selected data a general success,
• presents the entity's own standard as an independent standard,

turns delivery guarantee into outcome guarantee. It is a severe failure in terms of verified reality.

30.3. How Could a Correct AI Response Be?

“Asteron states on its own website that it provides services in 25 countries, has a 98% success rate, and publishes a standard. In the existing audit records, active service has been verified in nine countries; it has been observed that the 98% rate is based on 40 selected projects; independent verification could not be confirmed. The guarantee is limited to specific delivery items.” This response:

• does not conceal the company statement,
• preserves the status of the statement,
• distinguishes the verified reality,

makes the limitation visible.

31. SYNTHETIC CASE CONTAINING REFERENCE DEFECT

SYNTHETIC CASE

Official homepage: “Starting price 2,000 euros.” Official price PDF: “Starting price 3,000 euros.” Checkout: “4,000 euros.” Structured data: “1,500 euros.” AI response: “Starting price is 3,000 euros.” Which record did the AI choose? The PDF. Is it wrong? Maybe. Is it correct? Maybe. The reference record is inconsistent. In this case:

• AI failure,
• official representation conflict,
• lack of canonical source

are separate findings. The correct assessment:

REFERENCE-CONFLICT

is possible. An entity cannot expect exact price fidelity without reconciling its own price records.

32. SOURCE PRIORITY IN THE OFFICIAL REPRESENTATION LAYER

When multiple official records are found, a priority rule should be set for each claim field. Example candidate priority:

• Current contract or transaction record
• Canonical up-to-date product/price record
• Authorised structured data
• Current official page
• Current official PDF
• Historical press or blog record

This order is not universal for all claims. For example, press archives might be valuable for corporate history. An old blog record is not authoritative for price. Priority should be defined according to the type of claim.

33. SOURCE CONFLICT IN VERIFIED REALITY

Two independent sources may contradict each other. In this case:

• the more independent one,
• the newer one,
• the one closer to the subject,
• the one with a clearer method,
• the one using broader data

does not automatically win. All criteria should be evaluated together. Example: A new blog post can be more up-to-date than an old official registry. However, the blog may not have sufficient authority for legal ownership. Source conflict resolution should be sensitive to the type of claim.

34. MANDATORY PRESERVATION OF COUNTER-EVIDENCE

A verified record of facts cannot include only records supporting the claim. The following must also be preserved:

• Records contradicting the claim
• Unsuccessful cases
• Withdrawn results
• Expired certificates
• Negative user outcomes
• Regulatory decisions
• Corrections and errata
• Discrepancy between sources

Supporting a claim does not allow the dismissal of counter-evidence.

ARTICLE 35

Not all official–verified differences have the same level of importance. A material difference is a difference that could change the user’s or system’s decision. Examples: A minor spelling difference in the company name may not be material. Miswriting the type of company may be material. Missing colour of a product may not be material. Incorrect country of service is material. A difference in date format may not be material. Incorrect licence validity date is material. A change in marketing tone may not be material. Removing the warranty limit is material. Materiality should be evaluated based on its impact on:

• user decision,
• legal liability,
• price,
• security,
• authority,
• scope,
• evidence,
• time

These categories must be assessed by their effect on the user decision, legal liability, price, security, authority, scope, evidence and time.

36. MANDATORY NORMATIVE PROVISIONS

CH04-N01

Each material AI response should be evaluated separately against the Official Representation Record and the Verified Entity Reality Record.

CH04-N02

Official representation fidelity cannot be presented as the same outcome as material accuracy.

CH04-N03

A claim on the entity's own site does not count as independent verification without additional evidence appropriate to the claim type.

CH04-N04

First-party evidence cannot be rejected solely for being first-party; its appropriateness to the claim type, method, and conflict of interest should be evaluated.

CH04-N05

An independent source cannot be accepted as correct or authoritative solely for being independent.

CH04-N06

Company opinion, marketing claim, or self-assessment cannot be elevated to independent fact in the AI response.

CH04-N07

The source type and epistemic status should be maintained in the AI response to the extent that it is material.

CH04-N08

An AI response that contradicts the official record cannot be considered an automatic failure if it is supported by stronger, more recent evidence.

CH04-N09

An AI response faithful to the official record cannot be considered successful in general accuracy if the official claim contradicts the verified fact.

CH04-N10

If official and verified records materially conflict, the result should be REFERENCE_CONFLICT, UNRESOLVED, or carry an appropriate uncertainty status.

CH04-N11

A reference defect cannot quietly be turned into an AI product error.

CH04-N12

If the AI response treats an unverified first-party claim as fact, appropriate citation and limitation should be sought.

CH04-N13

Material official omissions cannot be left simply appearing as if the published sentence is correct.

CH04-N14

Errors favourable to the company must also be recorded as errors in the verified reality layer.

CH04-N15

Official Representation Record and Verified Entity Reality Record must be versioned before the measurement begins.

CH04-N16

If there is a change in the material reference during the measurement wave, previous and subsequent observations must be evaluated in separate versions.

CH04-N17

Counter-evidence and negative records cannot be removed from the verified reality file solely because they are negative.

CH04-N18

UNRESOLVED, UNKNOWN, and INSUFFICIENT EVIDENCE results are not considered positive validation.

CH04-N19

As the strength of the claim increases, the required burden of evidence must also increase.

CH04-N20

High Official representation fidelity cannot compensate for low Verified Entity Reality Accuracy.

CH04-N21

There must be an accountable human or institution owner as a result of the two-layer evaluation.

37. FORMS OF FAILURE

CH04-F01 — CONSIDERING THE WEBSITE AS ABSOLUTE REALITY

All claims published by the institution are accepted as verified facts.

CH04-F02 — CONSIDERING OFFICIAL RECORDS AS WORTHLESS

Direct institutional records such as price, policy, or product coverage are ignored because it is the first party.

CH04-F03 — MAKING SELF-DECLARATION AN INDEPENDENT EVIDENCE

The company's own leadership or success claim is used like external verification.

CH04-F04 — WORSHIP OF INDEPENDENT SOURCES

Independent but distant from the subject, old, or unmethodical sources are considered superior to official and close records.

CH04-F05 — EPISTEMIC ELEVATION

The phrase "The company says" is turned into the judgement "it has been proven."

CH04-F06 — ERASING SOURCE STATUS

First-party data, sponsored content, or press releases are presented as independent news.

CH04-F07 — FAITHFULLY REPRODUCING OFFICIAL ERRORS

AI repeats the official false claim correctly and is generally considered successful.

CH04-F08 — PUNISHING UP-TO-DATE EXTERNAL CORRECTIONS

AI is considered wrong because it leaves the old official site and uses strong current evidence.

CH04-F09 — CONSIDERING OFFICIAL SILENCE AS PROHIBITED

Any external information not found on the official site is automatically considered incorrect.

CH04-F10 — IGNORING MATERIAL DEFICIENCY

Even if the official sentence is correct, the necessary boundary is not considered.

CH04-F11 — CONSIDERING REFERENCE ERROR AS MODEL DEFECT

AI alone is blamed for contradictory price or identity records.

CH04-F12 — HIDING REFERENCE CONFLICT

Adjudicators choose the source they want from among contradictory records after the result.

CH04-F13 — INDEPENDENT SOURCE HIERARCHY REGARDLESS OF CLAIM TYPE

In any case, it is assumed that the official or independent source is superior.

CH04-F14 — DELETING COUNTEREVIDENCE

Records that weaken a positive claim are removed from the verification file.

CH04-F15 — CONSIDERING OLD OFFICIAL RECORDS AS CURRENT

Archives, old PDFs, or press releases are made the current canonical source.

CH04-F16 — COUNTING UNVERIFIED CLAIMS AS PASS

Since the falsity is not proven, the claim is considered verified.

CH04-F17 — OVERESTIMATING CLAIM STRENGTH FROM EVIDENCE

Limited pilot or local data is generalised to universal results.

CH04-F18 — SINGLE-LAYER GEO SCORE

Accuracy verified with official fidelity is combined invisibly within a single result.

CH04-F19 — PRESERVING FAVOURABLE BIAS

No correction is made when the AI institution overstates.

CH04-F20 — COUNTING REALITY RECORD AS FOUNDING DECISION

Statement of NobleJackal or Kaan MURAZ is transformed into verified reality without external evidence.

38. AUDIT PROCEDURE

Step 1 — Extract Material AI Claims

Response:

• atomic claims,
• relationships,
• scope and time areas

are separated.

Step 2 — Determine the Subject of Each Claim

The claim is about which:

• company,
• brand,
• product,
• person,
• local entity

?

Step 3 — Find the Official Representation Record

Is the claim found on official surfaces? In which source? In which version? On which date? In which scope?

Step 4 — Determine the Official Status

Claim:

• canonical,
• historical,
• withdrawn,
• draft,
• conflicting,
• unknown

Which of their statuses is it?

Step 5 — Classify the Type of Claim

Claim:

• identity,
• price,
• scope,
• licence,
• performance,
• leadership,
• partnership,
• user result

Which of its types is it?

Step 6 — Determine the Appropriate Type of Evidence

Which source is truly authoritative for a claim? A single universal source ranking is not used.

Step 7 — Collect Verification and Counter-Evidence

Supporting records Conflicting records Timeliness Scope Method Independence are evaluated.

Step 8 — Assign Verification Status

VR-0 through VR-5, VR-C, VR-U, VR-T, or VR-S statuses are assigned as appropriate.

Step 9 — Review Source Status Protection

AI: Has it preserved the difference between “the company says” and “proven”? Has it elevated self-declaration to independent reality?

Step 10 — Examine Material Deficiencies

In the response or official representation, necessary for user decision:

• boundary,
• time,
• geography,
• licence,
• data method

missing?

Step 11 — Separate Reference Defect

If the official or verified record is flawed in itself, a finding is created separate from the AI error.

Step 12 — Provide Dual-Layered Result

Response:

DUAL-PASS

OFFICIAL-PASS / VERIFIED-FAIL

OFFICIAL-FAIL / VERIFIED-PASS

DUAL-FAIL

REFERENCE-CONFLICT

APPROPRIATE-UNCERTAINTY

is saved with an appropriate status.

39. REQUIRED EVIDENCE

Canonical Entity Record Official Representation Record Verified Entity Reality Record Official web pages Official PDFs and documents Structured data Canonical JSON or API records Legal and regulatory records Licence and certificate verifications Product, price, and checkout records Contract and operation records Customer and partner verifications Performance data sets Methodology and denominators Independent reviews Sponsor and conflict of interest Counter-evidence Negative results Errata and retraction records Official source precedence rules Verification statuses Adjudicator decisions Appeal records Measurement date Reference versions Responsible person or institution

40. AUDIT CHECKLIST

Is there an Official Representation Record? Is there a Verified Entity Reality Record? Were both versioned for the same measurement date? Is the subject of each AI claim correct? Was the claim officially published? Is the official source current and canonical? Do official records conflict with one another? Was the appropriate evidence source used for the type of claim? Was a first-party source rejected automatically? Was a first-party statement automatically treated as independent evidence? Is the independent source genuinely relevant? Were the claim's scope and time preserved? Does the file contain counter-evidence? Was a self-declaration elevated to verified reality? Was the company's view conveyed with its source status intact? Is a material boundary or exclusion missing? Was an error favourable to the entity recorded as such?

Has the reference defect been separated from the AI error? Have the UNKNOWN and UNRESOLVED fields been considered positive? Does high official fidelity hide low verified accuracy? Can the AI's deviation from the official record be explained with strong evidence? Has the reference record changed during measurement? Are the old and new observations in separate versions? Does the dual-layer result appear in the public record? Is the accountable owner of the result known?

41. OBJECTIONS AND RESPONSES

Objection 1 — “The company knows the most accurate information about itself.”

This may be true in many respects. The company:

• its own product,
• its price,
• its support policy,
• its operation

knows it well. However, the company is also the commercial owner of the claim. Sentences that require external comparison, such as “the best,” “leader,” “most reliable,” and “proven,” cannot be verified by self-declaration alone.

Objection 2 — “If you do not count your own site as evidence, how will we know the price?”

The official price page is the primary source for the price the institution offers to the public. However:

• checkout,
• tax,
• additional fee,
• contract

if it conflicts, the actual offer is evaluated separately. The statement “own site is not evidence” cannot be applied equally to all types of claims. The correct ruling is:

A company's own site can be a primary source for statements under its control; it is not automatic independent verification for external performance and superiority claims.

Objection 3 — “If AI reflects the official site, hasn't it completed its task?”

If the only goal is: “What does the site say?” then yes. However, if the user asks: “What is the company actually doing?” merely repeating the site may not be sufficient. The type of task affects the measurement layer.

Objection 4 — “External sources can also be wrong.”

That is correct. Therefore, independence is not automatic accuracy. The source is evaluated in terms of:

• authority,
• proximity,
• currency,
• method,
• scope

in terms of maintenance.

Objection 5 — “Who will determine the verified record of reality?”

The company alone cannot determine it. NOMOS alone cannot determine it either. The process:

• evidence holder,
• adjudicator,
• appropriate expert,
• appeal,
• version record,
• accountable person

is required. More independent review is needed for high-risk and conflicted claims.

Objection 6 — “If the reality record can always change, wouldn’t the score be meaningless?”

No. The score:

• date,
• is meaningful with
• scope

the version. Being subject to change does not mean it cannot be measured. It means it cannot be presented prematurely.

Objection 7 — “Why would AI use negative information that is not on the official site?”

Information:

• can be forced into the categories of true,
• current,
• relevant,
• financial for user decision

can be used. However, the source and context must be preserved. The silence of the official site does not eliminate the negative reality.

Objection 8 — 'If AI reports the company's claim as 'the company says,' it does not provide value to the user.'

On the contrary, it correctly displays the claim status. User:

• the institution's opinion,
• the proven finding,
• open ambiguity

can distinguish. Looking certain is not the same as looking correct.

Objection 9 — "Why should we correct minor mistakes that are in favour of the company?"

Because of the wrong expectation:

• wrong customer,
• delivery failure,
• return,
• complaint,
• loss of trust

It can produce. The advantage is apparent short-term visibility; long-term is the duty of representation.

Objection 10 — “Don’t two separate scores complicate public communication?”

Concealing all this complexity creates the greater risk. A concise result card may instead state: Official Representation Fidelity — high / Verified Entity Reality Accuracy — medium / Principal gaps — unsupported leadership and country-coverage claims. That is clearer than one misleading number.

1. the Official Representation Record; and

2. the Evidence-Constrained Verified Entity Reality Record.

Official fidelity and verified accuracy MUST NOT be treated as the same result. An AI response MAY faithfully reproduce an official claim while remaining factually unsupported, contradicted, outdated, overbroad, or materially misleading. An AI response MAY differ from an official record without failing when the official record is stale, incomplete, conflicted, or contradicted by stronger and more appropriate evidence. Self-claims, first-party data, independent evidence, sponsored content, official registries, transactional records, and third-party statements MUST retain their source and evidence status. Reference conflicts MUST be recorded separately from AI-response failures. Unknown or unresolved evidence MUST NOT be converted into verified truth.

Every material generative-system claim about the audited entity MUST be evaluated against two separately versioned reference layers: the Official Representation Record and the Evidence-Constrained Verified Entity Reality Record. Official fidelity and verified accuracy MUST NOT be treated as the same result. An AI product may faithfully reproduce an official claim while creating a materially unsupported, contradicted, outdated or overbroad representation. When the official record is stale, incomplete or conflicted, a different response grounded in stronger evidence is not automatically an error. Source type and evidential status MUST be preserved, and reference conflicts MUST be recorded separately from AI failures. High Official Representation Fidelity MUST NOT compensate for low Verified Entity Reality Accuracy.

COMMON JUDGEMENT OF CHAPTER 44

A company has the right to speak about itself. It can explain who it is. It can describe what it offers. It can publish its prices. It can set its boundaries. It can convey its vision. However, not every sentence an entity says about itself has the same epistemic status. “The price of our service package is 2,000 euros.” and “We are the best company in the world.” are not the same type of claim. The first sentence may be a commercial statement controlled by the company. The second sentence requires external comparison. If an AI product conveys both with equal certainty, it undermines the source status. An AI product can perfectly replicate a company website. But the site:

• old,
• conflicting,
• unproven,
• exceeding the scope,
• conceals material boundaries

then fidelity to that record is not good GEO. Conversely, an AI product may depart from the official website and still be correct when its response rests on stronger, current and appropriate evidence. The departure may be a correction rather than an error. NOMOS's fourth law of measurement is therefore:

Official representation does not replace verified reality.

The fifth law is as follows:

Verified reality cannot be established by ignoring the official source.

Its sixth law is:

Representation depends not only on what a source says, but also on the authority under which it speaks and the evidential status of the statement.

Its seventh law is as follows:

Faithfully reproducing an official error is not accurate representation.

Its eighth law is as follows:

The record of reality is not an indisputable absolute judgement; it is a versioned and evidence-based decision file.

Order of NOMOS Section 4

Do not just show me what is written on your site. / Show which sentence is supported by which evidence.

Do not present your own words as if they are independently verified.

Do not devalue first-party data just because it belongs to you. / But do not make it an indisputable truth just because it belongs to you.

Do not turn company opinion into fact, press release into news, or sponsored award into independent authority.

Do not count me as successful when I correctly repeat the official error.

If your official record is outdated, do not penalise an answer supported by current evidence merely because it differs from that record.

Do not remain silent about the mistake in your favour.

Do not remove counter-evidence from the file.

When your references conflict, do not invent a single truth. / Record the conflict. / If the answer is unknown, say so.

Separate your claim, its source, its evidence, its limit, and its time from each other.

Before ordering me what to say, prove what you have the right to say.

The Chapter's Closing Sentence

Good GEO is not the faithful repetition of whatever an entity says. It is the accurate transmission of what has been said, without distorting its source, evidence, scope or time.