Method status

Author: Eugene Prudchenko Version: 1.0 Status: Developing methodology Framework: The Semantic Authority Network Empirical starting point: Case Zero: The Search Result Is Not the Page Conceptual stress test: Stress-Testing the Semantic Authority Network with Delphi Digital Minds Source boundary: The Delphi stress test used AI-generated responses from Delphi digital-mind profiles; it was not direct expert review, peer review, or endorsement. Applied diagnostic direction: EntityProof

1. Purpose

Search and AI systems do not form an understanding of a person or company from one page alone.

They encounter a distributed public environment: official websites, biographies, service pages, articles, profiles, directories, reviews, documents, interviews, structured data, citations, stale records, copied claims, and third-party discussion. They then select, combine, omit, or misinterpret parts of that environment for a particular query.

The Semantic Authority Methodology provides a disciplined way to examine that process.

Its purpose is to help a reviewer:

  1. define the entity under examination;
  2. separate claims from evidence;
  3. distinguish current information from historical information;
  4. map owned and independent sources;
  5. inspect machine-readable relationships;
  6. observe live retrieval and representation;
  7. adjudicate gaps without inventing certainty;
  8. apply bounded corrections;
  9. measure whether representations become more accurate, coherent, and useful.

The methodology does not promise rankings, citations, recommendations, or control over model output.

Its objective is narrower:

Build and evaluate a truthful source environment from which more accurate machine reconstruction becomes supportable.

2. Relationship to the Semantic Authority Network

The Semantic Authority Network is the primary public framework.

It describes the connected system of canonical identity, original concepts, inspectable evidence, owned sources, independent references, machine-readable relationships, and observed representations through which search and AI systems reconstruct an entity.

The Semantic Authority Methodology is the operating standard for applying that framework.

It defines:

Machine-readable authority for AI systems is not treated here as a separate framework. It describes a condition inside the Semantic Authority Network:

Machine-readable authority for AI systems is the condition in which a real entity’s identity, claims, evidence, and relationships are expressed consistently enough for machines to parse, compare, and represent them with bounded confidence.

EntityProof is the applied diagnostic direction built from this framework and methodology.

The Personal AI Layer remains a separate framework for human-governed personal context. Compound Operator describes the persistent human–AI working relationship enabled by that context. Neither is collapsed into the Semantic Authority Network.

3. Non-negotiable principles

3.1 Truth before visibility

The methodology may improve the visibility and legibility of real evidence.

It must not manufacture authority that does not exist.

Fake reviews, synthetic endorsements, invented metrics, hidden sponsorship, duplicated biographies, fabricated case studies, false affiliations, artificial consensus, and misleading profile networks are forms of evidence corruption.

3.2 Authority and visibility are separate variables

An entity may possess real expertise but remain difficult for machines to retrieve.

Another entity may be highly visible while having weak or unsupported evidence.

Visibility does not prove authority. Authority does not guarantee visibility.

They must be recorded and measured separately.

3.3 Retrieval, interpretation, verification, and selection are different stages

Being retrieved is not the same as being represented accurately.

Accurate representation is not the same as being trusted.

Trust is not the same as selection by a buyer, partner, bank, regulator, or procurement team.

A claim of improvement must identify the stage that changed.

3.4 Public evidence has a hard boundary

Public sources cannot reveal all operational reality.

They may not reveal private references, current capacity, implementation reliability, performance under stress, confidential failures, pricing, working-capital constraints, or internal controls.

Absence from the public network is not proof of absence in reality.

The methodology evaluates public legibility and observed representation. It does not certify the complete quality or risk of an entity.

3.5 Structured data cannot create truth

Canonical URLs, authorship markup, dates, breadcrumbs, schema, and machine-readable files can clarify visible relationships.

They cannot turn an unsupported claim into evidence.

Technical correctness is necessary in some contexts, but it is not a substitute for factual support.

3.6 Contradictions are evidence

Conflicting sources should not be silently smoothed into one preferred narrative.

A contradiction may indicate:

The task is to classify the contradiction, not hide it.

3.7 Insufficient evidence requires a fail-closed conclusion

When the available source-linked evidence is insufficient, the methodology does not issue a confident negative or positive conclusion.

It records:

Insufficient evidence for a material conclusion.

The target state is not maximum scoring or maximum eligibility.

The target state is:

Sufficient evidence acquired, with no unsupported material conclusion.

4. Units of analysis

The methodology works with ten record types.

4.1 Entity

The person, company, organization, product, project, method, or concept under examination.

Required fields may include:

4.2 Claim

A statement that may affect how the entity is understood or selected.

Examples include:

Claims must be recorded separately from the pages on which they appear.

4.3 Concept

A defined idea, method, distinction, framework, or recurring question associated with the entity.

A concept record should include:

4.4 Evidence object

A specific artifact that supports, qualifies, contradicts, or limits a claim.

Examples include:

4.5 Source

A location where a claim, concept, evidence object, or representation appears.

Sources are classified as:

The classification concerns control and provenance, not automatic truthfulness.

4.6 Relationship

A visible or machine-readable connection between records.

Examples include:

4.7 Query context

The exact question or task under which a system is tested.

A query record includes:

4.8 Representation

The machine-generated output presented to a user.

Examples include:

4.9 Observation

A preserved record of what appeared under defined conditions.

An observation is not a causal explanation.

It should include:

4.10 Intervention

A bounded change intended to improve truthfulness, clarity, evidence access, relationship clarity, or representation accuracy.

Every intervention must identify:

5. Evidence model

Evidence is recorded across separate dimensions rather than collapsed into one authority score.

5.1 Support state

Evidence collection summary
StateMeaning
UnsupportedNo source-linked evidence supports the material claim.
Self-assertedThe claim appears on an owned or controlled source without inspectable supporting evidence.
First-party evidencedThe entity provides inspectable evidence that directly supports the claim.
Independently corroboratedA source outside the entity’s control adds specific confirmation, evaluation, application, or context.
Formally verifiedA competent institutional or authoritative source verifies the claim within its scope.
ContestedCredible sources materially disagree or evidence conflicts.
UnknownThe available material is insufficient to classify the claim.

5.2 Temporal state

Evidence collection summary
StateMeaning
CurrentSupported as presently effective.
HistoricalAccurate for a defined earlier period.
SupersededReplaced by a later supported state.
UndatedNo reliable effective date is available.
Time-conflictedSources assign incompatible dates or current status.

5.3 Independence state

Evidence collection summary
StateMeaning
OwnedControlled by the entity.
Controlled third-partyHosted elsewhere but materially created or controlled by the entity.
IndependentCreated outside the entity’s control and adds its own information or evaluation.
InstitutionalProduced by a competent public, professional, regulatory, academic, or standards body within its remit.
UnknownControl or provenance cannot be established.

5.4 Confidence

Confidence is expressed as high, moderate, low, or withheld.

Confidence must be tied to a defined conclusion, not to the entity as a whole.

6. The diagnostic method

Step 0. Scope the decision context

Define what decision the review is intended to support.

Required questions include:

A vague scope produces a vague diagnosis.

If entity, market, category, or decision intent cannot be defined, the review fails closed.

Step 1. Resolve the entity

Create the canonical entity record.

Map:

The output is not a preferred biography. It is a time-aware entity map.

Step 2. Register material claims and concepts

Extract the claims that materially affect understanding or selection.

For each claim, record:

Concepts are recorded separately from promotional claims.

A term does not become a framework through repetition. It requires a definition, boundaries, practical consequences, and relationships to evidence or application.

Step 3. Qualify the evidence

Attach inspectable evidence objects to each material claim.

For each evidence object, assess:

Do not treat the existence of a third-party profile as independent verification of every statement on that profile.

Do not convert a screenshot of one output into a claim about a durable platform mechanism.

Step 4. Map the source and relationship network

Build the graph through which machines and people may reconstruct the entity.

The map should include:

The objective is not maximum link density.

The objective is truthful relationship clarity.

Step 5. Observe retrieval and representation

Create a bounded, pre-registered query matrix.

Queries may include:

For each observation, preserve:

One favorable answer is not treated as durable recognition.

One absence is not treated as proof of invisibility.

Step 6. Adjudicate the gap

Classify each material issue before proposing a change.

Possible adjudications include:

#### Entity state

#### Category state

#### Evidence state

#### Representation state

#### Attribution state

The reviewer should state what is known, what is inferred, and what remains unverified.

Step 7. Apply bounded interventions

Only intervene where the diagnosis identifies a controllable gap.

Permitted intervention classes include:

Every intervention must preserve truth and disclose material limits.

The methodology prohibits:

Step 8. Remeasure and version

Repeat the pre-registered query and source checks using the same adjudication rules.

Record:

A before-and-after difference does not by itself prove causation.

The output is a versioned observation, not a victory claim.

7. Core measurement dimensions

The methodology does not produce one opaque authority score.

It measures separate dimensions.

7.1 Entity clarity

Can a reviewer and a machine distinguish the entity from similarly named or related entities?

7.2 Temporal coherence

Are current and historical roles, categories, claims, and relationships correctly separated?

7.3 Claim support coverage

What proportion of material claims have evidence appropriate to their risk and scope?

7.4 Independent corroboration coverage

Which claims that require external support have qualifying independent corroboration?

Copied distribution does not count as independent corroboration.

7.5 Cross-source coherence

Do material sources agree on identity, category, role, dates, and core claims without erasing legitimate history or disagreement?

7.6 Retrieval coverage

Across the pre-registered query set, where is the entity retrieved, omitted, or displaced by better-documented alternatives?

7.7 Representation accuracy

When retrieved, is the entity described accurately, incompletely, incorrectly, or with unsupported certainty?

7.8 Attribution integrity

Are authorship, ownership, source, quotation, and entity-content relationships represented correctly?

7.9 Source quality

Are systems relying on canonical and evidence-bearing sources, or on weak, stale, copied, or ambiguous material?

7.10 Buyer verification readiness

Does the public evidence provide enough clarity for a buyer or reviewer to continue verification?

This dimension does not certify the entity or replace procurement, legal, financial, regulatory, or operational due diligence.

8. Standard outputs

A complete review may produce:

  1. Scope Record
  2. Canonical Entity Map
  3. Claim and Concept Register
  4. Evidence Register
  5. Source and Relationship Map
  6. Query Matrix
  7. Representation Observation Log
  8. Gap Adjudication
  9. Intervention Register
  10. Remeasurement Report
  11. Version and Change Log
  12. Public Evidence Package, where publication is justified and privacy permits

Not every engagement requires every public artifact.

Private evidence must remain private unless there is a lawful, necessary, consented, and appropriately redacted basis for publication.

9. Failure modes

The methodology is designed to detect several recurring failures.

Entity ambiguity

The system cannot confidently determine which person, company, product, or concept a source describes.

Temporal drift

Past and current states are collapsed into one timeless profile.

Representation drift

A system continues to reproduce a stale, incomplete, or materially distorted version of the entity.

Attribution drift

Identity, authorship, quotation, or content signals move across URLs or entities without a visible supported relationship.

Category drift

The entity is repeatedly placed in an adjacent, historical, or incorrect commercial category.

Evidence gap

A material claim is easier to find than the evidence required to support it.

Independent-authority gap

The network contains extensive self-description but little qualifying external corroboration.

Source-preference gap

A system retrieves weak secondary material while stronger canonical evidence exists but is not selected.

Public-evidence gap

The entity may possess real capability that is not publicly inspectable.

Synthetic-authority corruption

The network appears extensive because controlled or fabricated material imitates independence.

10. Public and private evidence boundary

The public network and private operational reality must not be confused.

A public methodology may establish that:

It may not disclose or infer private facts merely to make the network appear complete.

Private information may be used for internal adjudication where lawful and authorized, but the public conclusion must distinguish:

11. Research and causal discipline

The methodology separates observation from explanation.

A preserved result can establish what was displayed.

It cannot, by itself, establish:

Competing explanations should remain competing until evidence distinguishes them.

Negative results and null results must be retained.

12. Falsifiability

The Semantic Authority Methodology should be narrowed or rejected where it fails to provide useful distinctions, predictions, or interventions beyond simpler methods.

The framework is weakened if repeated controlled studies show that:

Useful tests include whether systems:

13. Versioning and governance

The methodology is versioned because its classifications and procedures may change with evidence.

A new version should identify:

Historical versions should remain accessible unless legal, privacy, or security reasons require restriction.

The author remains responsible for the questions pursued, the evidence accepted, the interpretations retained, and the conclusions published.

AI systems may assist with source comparison, extraction, challenge, testing, and drafting. They do not replace human responsibility for evidence, judgment, or publication.

14. Minimum valid application

A review may not claim to apply this methodology unless it records at least:

If these elements are missing, the work may still be useful, but it should not be represented as a complete Semantic Authority Methodology review.

15. Conclusion

Semantic authority is not created by volume alone.

It is not created by a schema field, a favorable screenshot, a repeated slogan, a directory listing, or a large number of controlled profiles.

A stronger source environment emerges when:

The methodology does not command machines to understand an entity correctly.

It provides a disciplined way to make correct understanding more supportable, incorrect representations more diagnosable, and changes more measurable.

The page is not the final unit of analysis. The entity, evidence, relationships, retrieval contexts, and observed representations must be examined as one versioned system.