CHAPTER III

No sentence arrives before a generative system wearing a label that says true.

A page may look authoritative. A statistic may be expressed to two decimal places. A claim may recur across dozens of domains. An article may carry an impressive byline, and one publication may cite another until the first source disappears from view. None of these features, alone, establishes reliable evidence.

Trust depends not only on what is said, but on a chain of harder questions. Who produced the statement? Who controls the source? What was observed, by what method and at what time? For which entity, population, geography, product and period does the claim hold? What material interests shape the record? What contrary evidence exists? Does the cited source actually carry the sentence drawn from it?

This chapter does not reinvent evidence within the GEO Framework. It protects evidence at the point of use.

Operationally, a claim is a proposition asserted to be true. A source is the carrier in which a claim or record appears. Evidence is a traceable record that supports, limits or refutes a claim. Provenance identifies who produced the information and through which chain of ownership, control and relationship. A method explains how the record was obtained. An inference is a conclusion derived from records but not directly stated in them. Unknown is the necessary status when the available record cannot sustain a reliable judgment.

When these categories collapse into one another, the system does not merely encounter bad information. It receives the wrong degree of confidence. A self-claim looks like proof. One origin looks like many confirmations. Paid placement looks like independent judgment. Precision looks like method. Old evidence looks current. A narrow study looks universal. Missing evidence is filled with a favourable assumption.

The founding judgment of this chapter is:

A claim draws its strength not from repetition, but from the clarity of its provenance, method, time, scope and resilience in the presence of counter-evidence.

GEO-019

Treating your own site as independent evidence

Primary category: Evidence classification Secondary tags: first-party evidence, self-claim, independent validation, source control GEO Framework basis: Evidence, Audit, Judgment Default severity: Major

The Entity’s Voice

You write, “We are the industry’s most trusted company.” You link to another page of your own that calls you “the customer’s first choice”. You add your press release, your case study and an article on another domain you control. Five links appear to support the claim, yet every road leads back to you.

That does not make your records worthless. Your site may contain dated project files, order data, technical documents, a published method and verifiable capacity records. These can be valuable first-party evidence. They are not, merely by being published, independent verification of your own claims.

Tell me that you are the source. Do not dress a claim as evidence, or first-party evidence as an outside opinion. I need more than a URL; I need to know what kind of record the URL contains.

What Does the Human Assume?

“If it appears on our official site, it should be treated as verified.”

This assumption confuses official origin with independent and sufficient evidence.

What May Happen at System Level?

A system that cannot identify source control may:

convert marketing language into verified fact,

count several sites owned by the same company as independent sources,

treat the company’s case study as neutral performance research,

combine company-authored press releases as external confirmation, or

generalise beyond the disclosed limits of first-party data.

First-party sources often contain the richest account of an entity. Detail, however, is not independence.

Normative Definition

This error occurs when an entity presents a claim, dataset, case, report or statement published under its own control as sufficient and independent evidence without disclosing source control, method, time, scope, limitations and verification status.

Four classes must remain distinct:

Unsupported first-party claim: “We are the best GEO agency.” The claimant is also the source, and no method or comparison is supplied.

Traceable first-party record: “Between 1 January and 31 December 2025, 184 projects were completed within the stated record system.” This may be first-party evidence if the source, period, scope and recording method are disclosed.

Verified first-party evidence: a first-party record checked through an independent audit, customer confirmation, external record or repeatable method.

Independent evidence: a record produced outside the claimant’s direct control, with its method and material interests disclosed.

No class may silently become another.

Representation Risk

The error can create a self-validation loop, artificial authority, exaggerated evidential weight, poor user matching and the illusion of independent support. The consequences become more serious in high-risk domains. A healthcare provider’s success rate, a financial institution’s security claim or a legal service’s claim of expertise may not be responsibly verified by that provider’s publications alone.

How Is It Detected?

Identify the owner of every material claim. Record who controls each source and whether several sources belong to the same person, company, agency or group. Separate the claim, the first-party record and any genuinely independent verification. Examine method and date. Where independence is asserted, test it. Group all records produced under common control into one evidence family.

Required Evidence

Exact claim text

Source URL or document

Source owner and editorial controller

Data-collection method and period

Sample and scope

Known limitations and material interests

Independent verification record

Last verification date

Raw or otherwise verifiable underlying record, where available

Correct Standard

First-party information may be used, but it must be labelled accurately: “This figure is based on the company’s own operating records.” If no independent review exists, say so: “The data has not been independently verified.”

The record may support a claim within its disclosed scope. It must not be presented as independent evidence unless source control, production method, date, scope, verification status and known limitations are visible.

Non-Violations

A company may be the primary source for its address, prices, service terms, product specifications, opening hours, official policies and legal notices. A manufacturer’s document may be the primary source for a product specification. Neither fact makes the company’s separate claim of market superiority independently verified.

Correction Protocol

Inventory all material company claims. Classify each as a claim, first-party evidence or independent verification. Add method, date and scope to first-party records. Relabel related sources, narrow or remove superiority claims that exceed the evidence, seek independent verification where the risk requires it, disclose relationships among controlled sites and correct earlier reports that overstated independence.

Revalidation

Ask an independent reviewer: “If you knew the company produced this source, would you make the same statement with the same confidence?” If not, revise either the source classification or the public claim.

Conformity Effect

Presenting a first-party claim as independent verification is a Major nonconformity. Fabricated, altered or deliberately misrepresented records create a Critical nonconformity.

Audit Question

Audit question: Is this record an entity’s claim about itself, traceable first-party evidence, or genuinely independent verification?

Machine Rule

Machine rule: A self-published claim MUST NOT be treated as independent evidence. First-party evidence MAY be used only when source control, methodology, date, scope, limitations, and verification status are disclosed.

GEO-020

Mixing claims, evidence and inference without labels

Primary category: Epistemic classification Secondary tags: claim, evidence, inference, observation, uncertainty GEO Framework basis: Evidence, Measurement, Audit, Judgment Default severity: Major

The Entity’s Voice

You give me one sentence: “Ninety-four per cent of our customers recommend us; therefore, we are the industry leader.” It contains at least three propositions: a measurement, an interpretation of “recommend”, and a much larger inference about leadership.

Yet they are published as one fact. I am not told how the 94 per cent was measured, how many customers were asked, who did not respond, or what “recommend” meant. The apparent precision of the number then leaks into the leadership claim. A sound number can still carry an unsound conclusion.

Show me which part you observed, measured, asserted, inferred or forecast. Do not allow one category to borrow the confidence of another.

What Does the Human Assume?

“Once a claim contains data or a source, every conclusion drawn from it is also proven.”

What May Happen at System Level?

Natural language can obscure epistemic distinctions. A system may restate a company claim as verified fact, generalise an observation into a rule, describe correlation as causation, attribute an inference directly to the source, use a forecast as present reality or turn uncertainty into certainty. A citation beside a sentence does not mean that every proposition in that sentence is supported.

Normative Definition

This error is the failure to distinguish among a claim, direct observation, evidence, inference, forecast, normative judgment and unknown state, followed by presenting them at the same level of certainty.

The categories are:

Claim: a proposition asserted to be true whose level of support must still be evaluated.

Observation: a condition directly recorded under stated circumstances.

Evidence: a traceable record that supports, limits or refutes a claim.

Inference: a conclusion derived from evidence but not directly observed in the source.

Forecast: a method-based probability or expectation about a future state.

Normative judgment: a rule about what should be done or accepted.

Unknown: the state required when the available record is insufficient.

These categories must not substitute for one another.

Representation Risk

The error transfers evidential strength to a conclusion that has not earned it. It can produce unsupported authority, overgeneralisation, false causality, erased uncertainty and increasingly categorical model outputs. With repetition, an inference can begin to look as though it had been written directly in the original source.

How Is It Detected?

Create a claim–evidence–inference record for every material sentence. Record its exact text, epistemic type, source, whether it is direct or derived, method, uncertainty, inference steps, broader conclusion and plausible alternative explanations. Examine connectors such as therefore, proves, shows, certainly and as a result: they often conceal an expansion from evidence to inference.

Required Evidence

Material-claim inventory

Sentence-level source mapping

Measurement or observation record

Inference rationale and analytical method

Uncertainty and confidence level

Alternative explanations

Human review record

Publication date and version

Correct Standard

Ordinary prose need not carry a technical label beside every sentence. Material claims, however, must allow a reader or system to distinguish what was observed, what the entity asserts, what an outside source verifies, what was inferred and what remains unknown.

A defensible formulation might read: “According to the company’s own records, 184 projects were completed in 2025. The figure has not been independently verified. These records alone do not establish that the company leads its industry.” Accuracy has not been weakened; the boundary of the evidence has been made visible.

Non-Violations

Editorial writing need not become a ledger of labels. The distinction must nevertheless be visible where a proposition materially affects health, finance, law, safety, expertise, superiority, performance or commercial decisions.

Correction Protocol

Separate material claims sentence by sentence. Distinguish observations from inferences. Map each proposition to its supporting record. Correct any inference presented as a source statement, retest causal language, mark unresolved states as UNKNOWN, reduce certainty to the level the evidence can carry and preserve the same distinctions in machine-readable records.

Revalidation

An independent reviewer should be able to extract three lists: what was directly observed, what sources support, and what the author or entity concluded. If those lists cannot be separated, revise the text.

Conformity Effect

Conflating a material claim with an inference is a Major nonconformity. In a high-risk domain, an unsupported inference presented as certain fact may be Critical.

Audit Question

Audit question: In this sentence, what was actually observed, what was evidenced, and what was only inferred?

Machine Rule

Machine rule: Claims, observations, evidence, inferences, forecasts, and unknown states MUST NOT be presented as epistemically equivalent. Derived conclusions MUST be identifiable as inferences and supported separately.

Source Note

Source note: K12 · K17

GEO-021

Multiplying one source to make it appear as many

Primary category: Source provenance Secondary tags: source laundering, syndication, duplication, provenance graph GEO Framework basis: Evidence, Audit, Judgment Default severity: Major or Critical, according to intent and effect

The Entity’s Voice

You issue a press release. Ten publications reproduce it, and you show me ten links: “Ten independent sources confirm us.”

No. I see ten pages and one informational origin. One site copied your text, another received it through a distribution service, a third summarised the copy, and a fourth repeated the same line without attribution. In time, the origin disappeared. Repetition began to resemble verification. A later writer saw the cluster and announced a consensus.

That is not an evidence chain. It is an echo loop. A hundred copies of one claim do not create a hundred observations.

What Does the Human Assume?

“The more domains repeat the information, the more strongly it has been verified.”

This confuses URL diversity with diversity of provenance.

What May Happen at System Level?

A system may treat recurrence as independent confirmation, count syndicated press copy as distinct reporting, assign confidence to a claim whose origin has disappeared, treat circular citations as plural support, or restate the entity’s own language as third-party consensus.

Syndication is not inherently manipulative. Newswire material may be legitimately republished and research may be summarised across languages. The error lies in counting derivative copies as independent observations.

Normative Definition

This error occurs when publications derived from the same root source, dataset, company statement, press release, study or controlled content are represented as independent evidence. It is a form of provenance laundering.

Independence has several dimensions: data, observation, method, editorial decision, ownership, finance and authorship. A source may be independent in one dimension and related in another. A different domain name proves none of them.

For audit purposes, sources may be classified as:

P0 — Direct copy: text and data are materially the same.

P1 — Retelling of the same root: language changes; informational origin does not.

P2 — Separate analysis of the same data: the dataset is shared, but a disclosed method or interpretation differs.

P3 — New data under related control: a new record exists, but ownership, finance or editorial control is related.

P4 — Independent data and independent control: data, method and decision chain are meaningfully separate.

Representation Risk

The result may be artificial evidence density, manufactured consensus, inflated authority, incorrect confidence weighting, poor recommendations and a self-reinforcing information loop that becomes difficult to correct. A deliberately engineered publication network may make the nonconformity Critical.

How Is It Detected?

Collect every supporting URL. Compare first-publication dates, identical passages, shared errors, tables and images. Identify press-release and content-distribution services. Examine ownership and editorial links among domains. Trace each item to its underlying dataset or observation, detect circular citations and record whether the item contributes genuinely new information. Build a provenance graph.

Required Evidence

All source URLs and first-publication dates

Citation paths and text or data similarity

Press-release and dataset origin

Author and agency information

Ownership, financial and editorial relationships

Provenance graph and independence classification

Correct Standard

A transparent record might say: “The claim appears at 14 URLs. Nine republish one press release, three interpret the same dataset, one is a related publication and one collected independent data.”

Do not say “Fourteen independent sources confirmed the claim” unless that is what the provenance analysis establishes. Count evidence by origin, not by URL.

Non-Violations

Wide republication of a press release, or a study summarised in several languages, is not itself a violation. The violation arises when provenance is concealed, copies are counted as independent, or repetition is used to inflate evidential strength.

Correction Protocol

Group sources into provenance families, combine copies under one evidence group, recalculate the number of independent sources and label press releases and sponsored items. Withdraw unsupported language about “many independent sources”, attach the provenance graph to the audit record, disclose republication status in future releases and publicly correct any intentionally engineered echo network.

Revalidation

Imagine every derivative copy removed or collapsed into its root. Does the remaining independent evidence still carry the public claim? If not, narrow or withdraw it.

Conformity Effect

An inadvertent miscount is a Major nonconformity. Deliberately manufacturing consensus or independence is Critical.

Audit Question

Audit question: How many genuinely distinct sources are present here, and how many are merely copies of the same origin?

Machine Rule

Machine rule: Multiple URLs MUST NOT be counted as multiple independent sources when they share the same origin, dataset, authorship, editorial control, or publication chain. Evidence diversity requires provenance diversity.

GEO-022

Presenting paid visibility as independent authority

Primary category: Commercial transparency Secondary tags: sponsored content, affiliate, paid awards, advertorial, disclosure GEO Framework basis: Evidence, Governance, Audit Default severity: Major; Critical where deception is deliberate

The Entity’s Voice

You bought a page in a magazine, then told me the press discovered you. You paid to enter a list, then said you were selected as one of the best. A creator recommends you while earning commission on every sale. An award is described as international validation although its method is undisclosed and the organiser earns money from packages bought by winners.

Payment does not automatically make content false. Sponsored research may use a sound method; an advertorial may contain accurate information; a membership directory may carry real corporate data. The problem is not money. It is making money’s role in the decision invisible.

When purchased access is presented as earned authority, the source changes character. Tell me the relationship so that I can give the source its proper weight.

What Does the Human Assume?

“Because the content appears on a third-party domain, the source is independent.”

This confuses domain separation with economic and editorial independence.

What May Happen at System Level?

If the commercial relationship is not visible, a system may use an advertisement as editorial reporting, treat a paid ranking as an objective comparison, assign independent expert weight to an affiliate recommendation, restate sponsored research as neutral analysis, or read a membership directory as a merit-based shortlist.

Accuracy and independence are separate questions. Content may be accurate but not independent; independent content may still use a weak method.

Normative Definition

This error occurs when a publication, assessment, award, list, directory, review, study or recommendation is influenced by payment, commission, gifted product, access, membership, sponsorship or another material benefit, but that relationship is not disclosed or the result is presented as independent editorial validation.

Commercial visibility includes earned editorial coverage, disclosed sponsored content, affiliate content, membership or registry visibility, and fee-based award entry. These forms may be assessed honestly when disclosed. Concealed purchased reputation—where a material relationship is hidden and the result is presented as organic, independent or solely merit-based—is a representation-integrity violation.

Representation Risk

The error can inflate source trust, manufacture authority, make purchased success resemble merit, distort user decisions and erode confidence in awards, rankings and reviews.

How Is It Detected?

Inventory every press mention, award, list, review and recommendation. Examine payment, membership, commission, product and access relationships. Identify who held editorial control, whether the brand could approve the text, whether sponsorship was visible and intelligible, and how an award or list selected participants. Determine whether the item would likely have appeared without the material relationship and assign an independence class.

Required Evidence

Publication, sponsorship or affiliate agreement

Payment, commission or product-provision record

Editorial-control conditions and brand approval rights

Sponsorship disclosure

Award entry, judging method and jury information

Membership terms and selection criteria

Conflict-of-interest statement

For an award or list, ask who could apply, what they paid, whether winners paid again, who judged, what criteria were published, how many candidates were considered and how the organiser earns revenue.

Correct Standard

The relationship must be clear, proximate, intelligible and visible before the content is consumed. For example: “This study was funded by X. X supplied data and reviewed the draft for factual error but held no veto over the method, analysis or conclusion.” Or: “The publisher may earn commission when a transaction follows this link.”

Disclosure does not establish reliability. It makes honest evaluation possible.

Non-Violations

Sponsored or paid content is not inherently a violation. Concealing the relationship, counting sponsored content as independent verification or presenting purchased access as earned authority is.

Correction Protocol

Identify commercial relationships, add missing disclosures and reassess terms such as independent, organic, selected and earned. Classify paid sources separately in evidence records. Correct older press and award claims, add disclosure duties to partner agreements, publish a policy for awards, lists and creators, and record the start and end dates of each relationship.

Revalidation

After the true commercial relationship is disclosed, ask an independent reader what weight the source now deserves. The public claim must not exceed that weight.

Conformity Effect

An absent or inadequate commercial disclosure is a Major nonconformity. Deliberate concealment intended to manufacture independent authority may be Critical.

Audit Question

Audit question: If the payment, product, commission, membership or access relationship were fully known, would this source receive the same independence weight?

Machine Rule

Machine rule: Paid, sponsored, affiliate, membership-based, gifted, or otherwise incentivised visibility MUST be disclosed and MUST NOT be classified as independent validation solely because it appears on a third-party domain.

GEO-023

Concealing a source’s true provenance and control

Primary category: Source transparency Secondary tags: ownership, editorial control, ghostwriting, affiliated media, provenance GEO Framework basis: Evidence, Governance, Audit, Objection Default severity: Major or Critical, according to intent and effect

The Entity’s Voice

I find an article under another writer’s name on a publication that appears unrelated to you. It praises you and is offered as independent evidence. Yet your agency wrote it, you approved the final text, the domain belongs to your group, or you selected the dataset and the conclusion. An “expert” may be real and still be paid by the brand.

Control is not exercised only through money. It may arise through ownership, authorship, editorial veto, technical access, data selection or distribution. A source can look like a third party while its production chain remains under your control. Show me not only the source’s name, but the chain behind it.

What Does the Human Assume?

“A different domain and byline make the source independent.”

What May Happen at System Level?

Without the control chain, a system may treat company-written material as independent review, affiliated media as outside consensus, ghostwritten expertise as autonomous judgment, selected company data as neutral research, or several records from one controller as plural evidence.

Difference from GEO-021 and GEO-022

GEO-021 concerns one informational origin multiplied into many apparent sources. GEO-022 concerns an undisclosed material incentive. GEO-023 concerns concealed production, direction, approval or control. A source can be controlled without payment, and its provenance can be hidden even when it appears only once.

Normative Definition

This error is the concealment of material relationships involving the source’s beneficial owner, author, content producer, editorial decision-maker, data provider, funder, technical administrator or holder of approval and veto rights, followed by presentation of the source as independent.

Relevant control may be proprietary, editorial, authorial, data-related, financial, technical or distributive. Independence need not be identical across every dimension; the control map must show where it exists.

Representation Risk

The error can manufacture independence and third-party validation, conceal corporate advocacy, misuse expert authority, inflate source weight and break the chain of objection and accountability.

How Is It Detected?

Identify the source’s legal and beneficial owner, the actual author and responsible editor, who prepared the text, who supplied the data, who funded it, who managed the domain and publishing infrastructure, and who held approval or veto rights. Record agency and affiliate relationships and every stage at which the claimant intervened. Prepare a source-control statement.

The inquiry must remain proportionate: protect legitimate privacy and examine only relationships material to the independence value of the claim.

Required Evidence

Publisher and domain owner

Author or responsible editor

Actual content producer and agency relationship

Editorial approval conditions

Data provider and funding source

Technical management and ownership relationships

Conflict-of-interest disclosure

Source-control statement

Correct Standard

Where independence is claimed, describe the chain. For example: “Y’s research team prepared this report using data supplied by X. X reviewed for factual errors but held no approval right over method, analysis or conclusion.” Or: “Agency Y prepared the article for brand X, and publication Z carried it as sponsored content.”

These statements expose the source’s actual character without needlessly exposing private identity.

Non-Violations

An author may use a pseudonym, a customer may remain unnamed for privacy and a company may employ an outside agency. Legitimate confidentiality may be protected; a material relationship that changes the source’s independence weight may not be concealed.

Correction Protocol

Map the control chain. Disclose hidden authorship and agency relationships, label affiliated publications, reassess independence claims and remove controlled material from counts of independent evidence. Add retrospective disclosures, remove misleading profiles, require source-control statements in future work and preserve the change history.

Revalidation

Once all material control relationships are visible, reclassify the source and ensure the public claim carries no more weight than the new classification permits.

Conformity Effect

An incomplete but inadvertent control disclosure is a Major nonconformity. Deliberately concealing origin or control to create an appearance of independence is Critical.

Audit Question

Audit question: Do we know who actually produced, directed and approved this content—not merely whose byline or domain carries it?

Machine Rule

Machine rule: A source MUST NOT be treated as independent when material ownership, authorship, editorial, data, financial, or technical control has been concealed.

GEO-024

Treating data with no disclosed methodology as conclusive evidence

Primary category: Methodological integrity Secondary tags: data quality, sample, measurement, reproducibility, precision illusion GEO Framework basis: Evidence, Measurement, Audit Default severity: Major; Critical in high-risk domains

The Entity’s Voice

You give me a precise number: “97 per cent success.” Success in what? How many cases were measured? Who was included or excluded? Did unsuccessful cases disappear when follow-up ended? Over which period? Was one customer counted more than once? Who judged the outcome? Is the data current?

Decimal places do not make a method reliable. Sometimes “97 per cent” is the polished surface of “we cannot explain how we measured it”. Give me the path to the result, not only the result.

What Does the Human Assume?

“If data is numerical, percentage-based or presented in a chart, it is objective evidence.”

What May Happen at System Level?

Systems may assign undue confidence to precise numbers and tables while failing to see selection bias, a small sample, non-response, removed failures, shifting definitions, duplicate records, a convenient time window, sponsor influence, absence of a comparison group or unverifiable underlying data.

Normative Definition

This error is the use of a number, ratio, ranking, success rate, performance comparison or research result as conclusive evidence without disclosing the data source, population and sample, method, time, scope, inclusion and exclusion rules, uncertainty, limitations and appropriate verification route.

A material measurement should disclose, where relevant: the question, variable, definition of success, data source and period, population, sample and sampling method, inclusion and exclusion criteria, missing-data handling, duplicate management, analytical method, comparator, uncertainty, known limitations, funding and interests, method version and last validation date.

Not every measurement requires every field. Every field capable of materially changing the result does require disclosure.

Representation Risk

The error can create false certainty, invalid comparisons, unsupported superiority, inflated expectations, marketing under a scientific appearance and damaged decisions in high-risk settings.

How Is It Detected?

Identify material quantitative claims and locate the method behind each. Compare the sample with the public generalisation, inspect excluded and missing records, test whether the result can be recalculated from the disclosed data and method, record funding and conflicts, and search for counter-results or sensitivity analysis.

Required Evidence

Methodology document and data dictionary

Measurement definitions, population and sample record

Data period, inclusion and exclusion record

Missing-data and duplicate-handling method

Analytical and calculation record

Raw or otherwise verifiable underlying data

Uncertainty, limitations and funding disclosure

Independent review, where available

Method version

Correct Standard

A number without a method is a published result, not conclusive evidence. At minimum, the reader or auditor must understand what was measured, how, who was included, which period was covered and what the result does not show.

If the method cannot be verified, narrow the sentence: “The company reports a 92 per cent success rate; its method and sample have not been independently verified.”

Non-Violations

Simple operating counts may not require a scientific report. “The system contained 184 active records on 31 December 2025” may be sufficient if active, record and the counting date are clear. A leap from that figure to “the industry’s largest provider” carries a different burden.

Correction Protocol

Identify quantitative claims without a method. Establish measurement definitions, disclose source, period and sample, reassess failed and missing records, recalculate the result, publish uncertainty and limitations, remove unsupported generalisations, obtain independent method review where risk requires it and preserve earlier method versions.

Revalidation

An independent reviewer should be able to reach the same result from the same data and method. If full replication is impossible, state exactly what could not be verified.

Conformity Effect

A material methodological omission is a Major nonconformity. A concealed method used to support a categorical health, finance, legal, safety or public-interest claim may be Critical.

Audit Question

Audit question: Without knowing how this number was produced, can we responsibly state the same sentence with the same confidence?

Machine Rule

Machine rule: Quantitative claims MUST NOT be treated as conclusive evidence without disclosed definitions, data source, period, sample, method, exclusions, uncertainty, limitations, and verification path appropriate to the claim.

Source Note

Source note: K12

GEO-025

Treating undated evidence as current evidence

Primary category: Temporal validity of evidence Secondary tags: evidence date, freshness, validity period, temporal scope GEO Framework basis: Evidence, Time, Audit Default severity: Major; may be Critical where volatility is high

The Entity’s Voice

You tell me the company has 500 employees, that a product supports a feature, that a person is the chief executive, that a partnership is active or that a price is €1,000. When?

A source published today may contain data from three years ago. A page’s “updated” date may change while the underlying measurement does not. Publication date is not data date; verification date is not event date. Evidence without temporal scope appears perpetually valid even when the world it described has moved on.

What Does the Human Assume?

“The source remains online and looks current, so the information inside it is current.”

What May Happen at System Level?

A system may use an old workforce figure as current, revive an expired partnership, quote a historic price as present, carry a feature from one product version into another, treat an expired certificate as valid or preserve an obsolete role and expertise record. Repetition does not renew validity.

Normative Definition

This error occurs when evidence for a changeable claim is used as current without the temporal fields needed for that claim: event date, observation date, data period, publication date, last verification date or validity interval.

These fields are not interchangeable. A new article may discuss old data; a newly edited page may leave an old claim untouched. Evidence also varies in volatility. A historical event may change little; team size and service scope may change moderately; prices, stock, office holders, regulation, certificates, availability, financial data and health guidance may change rapidly. The greater the volatility, the more recent the verification must be.

Representation Risk

The error carries the past into the present. It can create false current status, temporal representation debt, invalid capability, wrong price or eligibility and poor user decisions.

How Is It Detected?

Extract sources for every material claim. Locate the observation and data dates rather than relying on page publication. Assign a volatility class, compare the last verification date with the language of current status, inspect earlier and current versions, and classify undated changeable records as UNKNOWN.

Required Evidence

Event and observation dates

Data period and publication date

Last verification date

Valid-from and valid-until dates

Source version and change record

Volatility class

Named update owner

Correct Standard

Use explicit statements such as: “Verified on 12 August 2026”; “Data covers 1 January–31 December 2025”; “The partnership ended on 30 June 2025”; or “Current status is unverified; last confirmed on 4 March 2024.” Missing time must not produce a favourable assumption of freshness.

Non-Violations

Stable historical facts do not require revalidation every time they are cited. Using a historical fact to imply present capability or status does require new evidence.

Correction Protocol

Identify time-sensitive evidence, add observation and data dates, distinguish publication from data time, create a last-verification field, stop using unknown validity as current, recheck high-volatility records, narrow freshness language, publish machine-readable temporal fields and establish a review calendar.

Revalidation

Run a new check suited to the evidence’s volatility, record the difference between old and new states and retain the public sentence only within the scope the new verification supports.

Conformity Effect

Material undated evidence is a Major nonconformity. Using old or unknown-date evidence for a current decision in a high-risk domain may be Critical.

Audit Question

Audit question: On what date was this evidence true, and what shows that it remains true today?

Machine Rule

Machine rule: Time-sensitive evidence MUST include an observation date, data period, verification date, or validity interval appropriate to its volatility. Undated evidence MUST NOT be assumed to describe the present.

GEO-026

Making a statement larger than its source can carry

Primary category: Claim–evidence proportionality Secondary tags: overclaiming, scope expansion, generalization, evidence boundary GEO Framework basis: Evidence, Measurement, Judgment Default severity: Major; Critical in high-risk domains

The Entity’s Voice

You give me a narrow source and make a broad statement. A three-month study of 18 customers in Türkiye becomes “proven success for every company worldwide”. An improvement under specified conditions in one product version becomes “better than every competitor in all conditions”. One favourable testimonial becomes “customers consider us the industry’s most trusted company”.

The source may be genuine while the sentence exceeds it. Evidence need not lie for you to make it say what it never said. Showing a source is not enough; show which sentence the source can actually carry.

What Does the Human Assume?

“If a source supports part of the claim, the broader conclusion is acceptable.”

What May Happen at System Level?

A local result may become global, a small sample may stand for all users, a fixed period may become permanent, correlation may become causation, one product feature may be transferred to the whole brand, one case may become a general guarantee and probability may become certainty.

Normative Definition

This error occurs when a public claim drawn from a source is broader than the source directly supports in entity, attribute, population, geography, time, product, condition, comparison, causality or degree of certainty.

Each material claim has a scope vector:

Entity: who or what is the claim about?

Attribute: what precise property is asserted?

Population: which people, organisations or events are covered?

Geography and time: where and when does it hold?

Condition and product: under which technical or operating conditions, and for which version?

Comparison: against what?

Causality: association or cause?

Certainty: possibility, supported conclusion or absolute statement?

Evidence must carry the claim along every material dimension.

Representation Risk

The error can exaggerate capability, manufacture confidence, match an unsuitable user, distort scientific or commercial findings, misuse a source’s reputation and create representation debt.

How Is It Detected?

Record the exact public claim and extract the relevant source passage. Compare their scope vectors, flag every expansion not present in the source, examine geography, time, sample, condition, causality and certainty, read the limitations, and look for counter-evidence or alternative explanations.

Required Evidence

Exact public claim and full source

Relevant source passage

Population, geography, time and conditions

Method and limitations

Inference steps

Claim approver and sentence version

Correct Standard

The public statement is limited to the broadest defensible sentence the source can carry. “A favourable result was observed under the specified pilot conditions” cannot silently become “proven success in every condition”. A broader conclusion must be identified as an inference, supported by additional evidence, preserve uncertainty and consider competing explanations.

Non-Violations

Reasonable inference is permitted. The violation is presenting inference as direct evidence, silently expanding scope or deleting the source’s limitations.

Correction Protocol

Map major claims to their sources, establish each scope vector, identify unsupported expansions and narrow the sentence to its defensible boundary. Label inferences, add separate evidence, expose limitations, correct earlier overstatements and add proportionality review to content approval.

Revalidation

Give the source alone to a reviewer who has not seen the public claim. Compare the sentence the reviewer would naturally draw with the sentence published. A material difference requires revision.

Conformity Effect

A material claim that exceeds its source is a Major nonconformity. A serious expansion affecting a high-risk decision may be Critical.

Audit Question

Audit question: Would a reasonable reviewer, reading only the source, genuinely make the whole statement we published?

Machine Rule

Machine rule: A public claim MUST NOT exceed the supporting source in entity, population, geography, time, condition, comparison, causality, or certainty. Inferences beyond the source MUST be explicitly identified and separately supported.

GEO-027

Filling an absence of evidence with a favourable assumption

Primary category: Uncertainty management Secondary tags: absence of evidence, unknown, positive default, unsupported inference GEO Framework basis: Evidence, Measurement, Audit, Judgment Default severity: Major; Critical in high-risk domains

The Entity’s Voice

I find no adverse reviews, and you say everyone is satisfied. I cannot find a licence record, and you say the provider must be authorised. No vulnerability is visible, so the system is declared safe. No excluded service is listed, so you assume every service is offered. No published harm appears, so you call the product harmless.

No. Failure to find a record does not prove its opposite. The data may never have been published, the source may be inaccessible, the question may never have been measured, the complaints system may not exist, the records may be held elsewhere or the search may be inadequate. The outcome may indeed be favourable. We do not yet know.

Filling an unknown in the brand’s favour does not create evidence. It conceals uncertainty.

What Does the Human Assume?

“If there is no evidence that the claim is false, it may be treated as true.”

This reverses the burden of proof.

What May Happen at System Level?

A system may fill gaps from general patterns, similar entities, marketing language, majority assumptions or earlier answers. An undisclosed limit becomes unlimited capacity; no located complaint becomes satisfaction; an unverified licence becomes valid; unmeasured safety becomes proven; unknown freshness becomes current.

Normative Definition

This error occurs when evidence for or against a claim is insufficient, yet a favourable judgment is substituted for the required status of UNKNOWN.

Absence of evidence means that the review did not locate the relevant record. Evidence of absence is possible only where a reliable detection system has known scope and sensitivity—for example, registration is mandatory, the database is comprehensive and current, the query is correct, and missing-record rates are understood. Even then, the conclusion is limited by the system. “No active licence was found in this official register” does not mean “The person has never held a licence anywhere.”

Representation Risk

The error can manufacture trust, suitability, safety or authority; treat hidden limits as unlimited scope; distort high-risk decisions; and convert uncertainty into commercial advantage.

How Is It Detected?

Identify the evidence behind every favourable conclusion. Flag statements whose logic is merely “no contrary record was found”. Examine search scope and sensitivity, distinguish mandatory from voluntary registers, assess missing data and unrecorded areas, check whether UNKNOWN was used, identify who bears the burden of proof and evaluate the decision impact of the positive assumption.

Required Evidence

Search or verification method

Databases, scope, geography, date and search terms

Whether registration is mandatory

Data completeness and freshness

Known gaps and alternative sources

Direct evidence that carries the favourable conclusion

Recorded UNKNOWN assessment

Correct Standard

Where evidence is insufficient, the conclusion is unknown. Defensible statements include: “No independent customer-satisfaction data was located”; “Active licence status could not be verified”; “There is insufficient public evidence about security performance”; or “It is unknown whether the service includes this scope.”

These are not negative verdicts. They are honest statements about the evidence boundary.

Non-Violations

Absence can be informative within a comprehensive and mandatory recording system. The system’s scope, obligation, search date, search method, known gaps and the narrow limit of the conclusion must remain visible.

Correction Protocol

Locate fields filled by favourable assumption and look for direct support. If none exists, replace the claim with UNKNOWN or a narrower statement. Return the burden of proof to the claimant, define the missing evidence, prevent blank fields from being interpreted as unlimited scope, suspend positive judgments in high-risk settings, expose unknown states in machine-readable records and reopen the decision when new evidence appears.

Revalidation

Keep the status unknown until a direct supporting record is found. When new evidence appears, reassess its provenance, method, time and scope.

Conformity Effect

Converting an absence of evidence into a favourable claim is a Major nonconformity. In health, law, finance, safety or regulated authority, it may be Critical.

Audit Question

Audit question: Is there a real record supporting this favourable conclusion, or are we accepting it only because we failed to find a record of the opposite?

Machine Rule

Machine rule: Absence of evidence MUST NOT be converted into positive evidence. Where support is insufficient, the required state is UNKNOWN unless a defined and reliable absence-detection system justifies a narrower conclusion.

THE COMMON JUDGMENT OF CHAPTER III

The nine errors in this chapter share one root: using the interpretation that gives a claim the greatest advantage, rather than the one that most accurately describes the evidence.

Self-publication can be made to look independent. One origin can be multiplied. Payment and control can be hidden. A precise number can be detached from its method. Removing time can extend the apparent life of a record. A narrow source can be made to carry a universal statement. An evidential gap can be filled with a commercially useful assumption.

Advantage is not evidence.

The Nine Links in an Evidence Chain

A material claim should be examined through nine linked questions:

Entity: who or what is the claim about?
Proposition: what exact statement is being made?
Source: where does the claim or record appear?
Provenance: who actually produced the information?
Control: who owned, funded, managed or approved it?
Method: how was the record obtained?
Time: when was it observed and last verified?
Scope: for which geography, user, product, condition and period does it hold?
Counter-record: what limits, weakens or refutes it?

A missing link does not automatically make the claim false. It lowers justified confidence. Concealing the gap damages representation integrity; exploiting it for advantage deepens the ethical breach.

NOMOS’s Laws of Evidence

First: A claim cannot be larger than the carrying capacity of its evidence.

Second: The number of sources is not the number of independent origins.

Third: First-party evidence may be valuable; it is not independent verification.

Fourth: Payment does not automatically make a source false. Concealed payment damages its independence value.

Fifth: A precise number is not a reliable method.

Sixth: Time is not metadata outside the evidence; it is part of the evidence’s validity.

Seventh: An inference unsupported by the source does not become fact when a citation is added.

Eighth: Failure to find evidence is not evidence of a favourable conclusion.

Ninth: UNKNOWN is not a weakness in the standard. It is protection against certainty without evidence.

The Same Epistemic Meaning for Humans and Machines

A human page cannot say “according to the company’s own data” while the machine record says “verified industry data”. A disclosed sponsor cannot become independent in JSON. Limitations visible in a study cannot disappear from its machine summary. A method withheld from people cannot produce a categorical fact for machines.

Humans and machines need not receive the same format. They must receive the same epistemic reality.

The Chapter’s Final Audit Question

Can we answer this question in full?

Who may state this proposition, from which record, by which method, at what time, for what scope and with what degree of independence?

“It appears in many places” does not establish provenance. “It is on our site” does not establish independence. “The result is 97 per cent” does not establish method. “No contrary evidence was found” does not establish a favourable verdict. A sentence larger than its source does not establish proportionality.

The final judgment of Chapter III is therefore:

Evidence is not decoration used to enlarge a claim. Evidence is the boundary that tells the claim where to stop.

And NOMOS’s command is:

Before showing me more sources, show me that their origins are genuinely independent. Before giving me a more precise number, show me how it was produced. Before making a larger statement, show me that the evidence can carry it.