TOPICALAUTHORITY.ORG TAO / ROOT

Corroboration and Triangulation

TOPICALAUTHORITY.ORG / EVIDENCE SYSTEM / EVD-06INDEPENDENCE CONTROL ACTIVE
EVD / 06 · INDEPENDENCE + CONVERGENCE + COVERAGE

Agreement is cheap. Independence is evidence.

Corroboration occurs when genuinely independent evidence supports the same bounded claim. Triangulation strengthens interpretation by approaching that claim through different sources, methods, observation modes or times—without hiding divergence.

01 / CLAIMFIX THE PROPOSITION
02 / LINEAGETRACE UPSTREAM
03 / INDEPENDENCEREMOVE DUPLICATES
04 / AGREEMENTCOMPARE OBSERVATIONS
05 / TRIANGULATEDIVERSIFY PATHS
06 / QUALIFYREPORT CONVERGENCE
EVD / 06.1 DEFINITION

Multiple records are not necessarily multiple evidence lines.

Corroboration requires agreement plus informational independence. Triangulation adds diversity of access, method, measurement or time so one hidden failure is less likely to explain the entire result.

Count independent generating paths, not citations, domains, mentions or search results. Then test whether those paths converge on the same claim within compatible scopes.

Two sources can agree because one copied the other, both used the same dataset, both repeated a press release or both inherited the same measurement error. Agreement becomes corroborative only after shared dependencies are identified and bounded.

CORROBORATIONIndependent evidence agrees on the claim.
REPETITIONSeveral records repeat one upstream assertion.
TRIANGULATIONDifferent paths test the same proposition.
EVD / 06.2 CORROBORATION NETWORK DECODER

Four sources can mean one, two or four evidence lines.

Switch scenarios to see how lineage, method and time change the evidential meaning of apparent agreement. The visual count and the independent-line count are deliberately kept separate.

NETWORK / CLAIM C-042 · PROVIDER-NEUTRALINDEPENDENT CONVERGENCE
SOURCE / ADirect observationL1 · M1 · T0
SOURCE / BIndependent datasetL2 · M2 · T0
SOURCE / CDocument recordL3 · M3 · T0
SOURCE / DExpert analysisL4 · M4 · T0
CLAIM / C-042SUPPORTED
UPSTREAM / U1NONE SHARED
EVD / 06.3 INDEPENDENCE FINGERPRINTS

Independence has more than one failure point.

Two records can have separate publishers yet share data, method or institutional control. A robust fingerprint tests each dependency rather than applying a single yes/no label.

FINGERPRINT / SOURCE

Origin independence

Were the observations generated by different upstream sources rather than republished from one account?

LINEAGEL1 ≠ L2
CREATORSEPARATE
CONTROLNO SYNDICATION
TEST / WHO GENERATED IT?
FINGERPRINT / DATA

Dataset independence

Do sources observe separate data, or do they analyze the same hidden or licensed dataset?

INPUTD1 ≠ D2
SAMPLESEPARATE
CONTROLTRACE INPUTS
TEST / WHAT FED THE RESULT?
FINGERPRINT / METHOD

Method independence

Could one shared instrument, extraction rule or model reproduce the same error across sources?

METHODM1 ≠ M2
ERRORNON-SHARED
CONTROLDIVERSIFY MODE
TEST / HOW WAS IT OBSERVED?
FINGERPRINT / CONTROL

Decision independence

Are the sources governed, funded or editorially controlled by the same actor or incentive?

OWNERO1 ≠ O2
INCENTIVEINSPECTED
CONTROLDISCLOSE TIES
TEST / WHO CAN SHAPE IT?
EVD / 06.4 TRIANGULATION CUBE

Triangulation diversifies the path to the claim.

Source count alone increases repetition. Strong triangulation deliberately changes at least two dimensions while holding the claim boundary stable.

SOURCE DIVERSITYMETHOD DIVERSITYTEMPORAL DIVERSITY
AXIS / SSource triangulation

Different actors or systems with separately resolved lineages observe the claim.

AXIS / MMethod triangulation

Direct observation, documentary evidence, structured data and expert analysis test different failure modes.

AXIS / TTemporal triangulation

Repeated observations test whether support is stable, episodic, decaying or superseded.

CONTROL / CClaim invariance

The proposition, scope and terms must remain compatible or apparent convergence becomes a category error.

EVD / 06.5 AGREEMENT × INDEPENDENCE

Agreement and independence are separate coordinates.

The two-by-two model prevents duplicate consensus from being mistaken for corroboration and prevents genuine disagreement from being hidden.

Q1 / AGREE + INDEPENDENT

Corroborative convergence

Independent evidence lines support compatible versions of the same bounded claim.

ROUTE / INCREASE SUPPORT, KEEP LIMITS
Q2 / AGREE + DEPENDENT

Repetition

Several records agree because they share an upstream source, dataset, method or control.

ROUTE / COLLAPSE TO ONE LINEAGE
Q3 / DISAGREE + INDEPENDENT

Material divergence

Independent observations conflict. Scope, time, method and source quality require diagnosis.

ROUTE / EVD-07 CONFLICT RESOLUTION
Q4 / DISAGREE + DEPENDENT

Unstable lineage

Versions from the same evidence family contradict one another or mutate across republication.

ROUTE / TRACE VERSION + TRANSFORMATION
EVD / 06.6 CONVERGENCE MATRIX

Confirmation strength depends on what remains shared.

The matrix keeps source, input, method, time and control dependencies visible. “Independent” is never inferred from different domain names alone.

EVIDENCE SET ↓ / TEST →
SOURCE LINEAGE
INPUT DATA
METHOD
TIME
EVIDENTIAL STATE
Four syndicated articlesONE ORIGINAL REPORT
SHARED

All trace to U1.

SHARED

Same reported facts.

REPUBLICATION

No new observation.

NEAR-IDENTICAL

Publication lag only.

ONE EVIDENCE LINE

Agreement without corroboration.

API + direct observationSEPARATE ACCESS PATHS
SEPARATE

Different origin paths.

PARTLY SHARED

Same target system.

DIVERSE

Structured vs direct.

ALIGNED

Compatible window.

METHOD CORROBORATION

Shared target is intentional.

Three reports, one datasetSEPARATE ANALYSIS
SEPARATE AUTHORS

Distinct publications.

SHARED D1

Same upstream data.

PARTLY DIVERSE

Different analysis rules.

ALIGNED

Same data period.

QUALIFIED

One data error can affect all.

Repeated independent capturesMULTIPLE WINDOWS
SEPARATE EVENTS

New acquisition each time.

SAME TARGET

Expected dependency.

SAME METHOD

Method error persists.

DIVERSE

T0, T1 and T2.

TEMPORAL SUPPORT

Stability, not method diversity.

EVD / 06.7 WORKED EVIDENCE SETS

Corroborate the proposition—not the surrounding narrative.

Each set keeps its claim narrow and states exactly which independent path contributes new support.

SET / 01SEARCH OBSERVATION

A result state appeared for a query

A structured API response and a separately captured browser observation agree on the URL, position, locale and time window.

LINE ASTRUCTURED RESPONSEM1 / L1
LINE BDIRECT BROWSER CAPTUREM2 / L2
SHAREDTARGET SEARCH SYSTEMEXPECTED
SUPPORTS OBSERVED RESULT STATE — NOT RANKING CAUSE
SET / 02ENTITY IDENTITY

An organization controls a domain

The organization’s own declaration is matched with registry data and a technically observed domain relation.

LINE AFIRST-PARTY DECLARATIONSOURCE
LINE BREGISTRY RECORDDOCUMENT
LINE CTECHNICAL OBSERVATIONSYSTEM
THREE MODES TEST DIFFERENT PARTS OF THE RELATION
SET / 03AI RETRIEVAL

A source is repeatedly cited

Separate runs, prompts and retrieval contexts show repeated inclusion, while preserved citation targets verify that the system points to the same source identity.

LINE ARUN FAMILY / T0TEMPORAL
LINE BPROMPT FAMILY / P2METHOD
LINE CCITATION TARGET CHECKDIRECT
SUPPORTS OBSERVED RETRIEVAL PATTERN — NOT UNIVERSAL VISIBILITY
SET / 04FALSE MULTIPLICITY

Twenty pages repeat one statistic

All citations terminate at one undated report, and several pages copied each other without inspecting the original dataset.

RECORDS20 PUBLIC URLSVISIBLE
LINEAGEONE UPSTREAM REPORTU1
DATAUNRESOLVEDGAP
TWENTY MENTIONS COLLAPSE TO ONE UNVERIFIED LINE
EVD / 06.8 FALSE CORROBORATION

Consensus can be manufactured by the network itself.

These patterns create the appearance of confirmation without adding an independent observation.

FAIL / 01Syndication count

Republished copies are counted as independent sources.

COLLAPSE / SHARED ORIGIN
FAIL / 02Citation cascade

Later articles cite summaries that ultimately point to one assertion.

TRACE / TERMINAL SOURCE
FAIL / 03Shared dataset blindness

Different analysts inherit the same undisclosed sampling or collection error.

TRACE / INPUT LINEAGE
FAIL / 04Shared method error

Separate sources use one instrument, model or rule with the same bias.

TRACE / METHOD DEPENDENCY
FAIL / 05Same-time illusion

Simultaneous observations capture one temporary event and are generalized as stable.

ADD / TEMPORAL DIVERSITY
FAIL / 06Scope slippage

Sources agree on narrower claims but are cited for a broader proposition.

REPAIR / CLAIM BOUNDARY
FAIL / 07Majority substitution

Source count replaces inspection of competence, access and provenance.

REPAIR / QUALITY PROFILE
FAIL / 08Divergence deletion

Conflicting observations are removed until only agreement remains.

ROUTE / EVD-07
EVD / 06.9 CORROBORATION PROTOCOL

Six controls turn source count into evidence structure.

The protocol reports both convergence and dependence, then routes unresolved disagreement forward rather than smoothing it away.

01

Bind the claim

Fix proposition, scope, entities, units and observation window.

OUTPUT / CLAIM C
02

Resolve provenance

Trace every record to its originating source, dataset and version.

INPUT / EVD-03
03

Collapse dependencies

Group syndicated, derived and common-input records into evidence families.

OUTPUT / LINEAGES L1…LN
04

Compare observations

Test compatible claims, terms, timing and result direction.

OUTPUT / CONVERGENCE MAP
05

Triangulate gaps

Add an observation path that changes source, method or time.

OUTPUT / DIVERSE CONTROL
06

Report the state

State independent count, shared limits, convergence and divergence.

NEXT / EVD-07 OR EVD-09
EVD / 06.10 UNIT OF CORROBORATION

Define the claim before counting support.

Corroboration can be established at the level of an exact proposition, not at the level of a document, brand article or whole narrative. A source may support one proposition, remain silent on another and contradict a third. Claim-level separation prevents surrounding agreement from being used to decorate an unsupported conclusion.

CLAIM PART / SUBJECTWhich exact entity or version?

Resolve names, aliases, subsidiaries, products, domains and historical versions. Independent about one corporate group does not automatically support claims about every brand or product it controls.

CONTROL / CANONICAL IDENTITY
CLAIM PART / RELATIONWhat is being asserted?

“Uses,” “owns,” “partners with,” “mentions” and “depends on” are different relations. Similar vocabulary must not collapse them into one proposition. Preserve modality and negation.

CONTROL / EXACT RELATION
CLAIM PART / VALUEWhich state, number or category?

Record the value together with its unit, denominator, category rule and tolerance. Two sources can agree directionally while disagreeing materially on magnitude.

CONTROL / MEASURE SPECIFICATION
CLAIM PART / SCOPEFor which population and boundary?

Geography, audience, device, market segment, query form and inclusion rules define where support applies. Narrow evidence supports a narrow claim unless an additional inference justifies extension.

CONTROL / EXPLICIT UNIVERSE
CLAIM PART / TIMEAt which valid moment or interval?

Separate publication, capture and effective dates. Several records may agree historically but fail to establish the current state of a volatile subject.

CONTROL / VALIDITY WINDOW
CLAIM PART / CONFIDENCEHow strongly does the record support it?

Direct observation, derived estimate and expert interpretation provide different support relations. Agreement between weak inferences does not become direct proof through repetition.

CONTROL / SUPPORT TYPE
corroboration unit = normalized claim + declared boundary + evidence lines

A document count answers how many containers were found. An evidence-line count answers how many sufficiently independent observation paths support the same bounded proposition. These are different quantities and must remain visibly separate.

COUNTIndependent paths to the claim
DO NOT COUNTMirrors, quotations and summaries as new observations
REPORTShared assumptions that survive diversification
EVD / 06.11 LINEAGE RECONSTRUCTION

Different URLs may lead back to the same observation.

Lineage reconstruction identifies where information originated, how it moved and whether later records add new evidence or only new presentation. The goal is not to punish citation; it is to avoid confusing distribution with independent confirmation.

01

Extract the claim

Capture the exact sentence, number, table entry or observed state that appears to support the proposition.

OUTPUT / CLAIM FRAGMENT
02

Follow citations

Trace links, footnotes, quoted experts, embedded charts and named datasets to the earliest recoverable evidence record.

OUTPUT / ORIGIN CANDIDATE
03

Compare wording

Repeated phrasing, identical rounding and shared errors can expose copying even when an explicit citation is absent.

OUTPUT / DERIVATION SIGNAL
04

Compare data

Inspect sample, timestamps, categories, missing values and transformation rules to determine whether two outputs share an upstream dataset.

OUTPUT / DATA FAMILY
05

Map control

Identify common ownership, funding, editorial control, infrastructure or analytical authorship that could create correlated error.

OUTPUT / CONTROL DEPENDENCY
06

Collapse echoes

Group derivative records into one lineage for corroboration counting while preserving each record’s distribution role.

OUTPUT / EVIDENCE FAMILIES
VISIBLE NETWORK

12 publications repeat the same statistic.

Twelve retrieval results may improve discoverability and show how a claim spread. They do not establish twelve independent measurements.

VISIBLE RECORDS / 12
EVIDENCE NETWORK

One survey generated the statistic.

After lineage collapse, the claim has one originating evidence line plus eleven derivative distribution records. Its weight depends on the survey, not the echo count.

INDEPENDENT LINES / 1
EVD / 06.12 INDEPENDENCE MODEL

Independence is multidimensional—not a publisher checkbox.

Two sources can be independent in authorship while dependent in data, method or institutional incentives. Record each dimension separately. The weakest shared dependency may explain the entire pattern of agreement.

DIMENSION / ORIGINSeparate primary observations

Did each evidence line collect or generate its own record, or does one line derive from another? Independent publication is not independent origin.

TEST / UPSTREAM SOURCE
DIMENSION / DATASeparate underlying datasets

Two analyses of the same dataset can test analytical robustness, but they do not provide independent sampling of the underlying reality.

TEST / SHARED INPUTS
DIMENSION / METHODDifferent error mechanisms

Survey, transaction record, direct capture and registry document fail differently. Method diversity is valuable when the methods genuinely observe the same claim.

TEST / FAILURE CORRELATION
DIMENSION / INSTRUMENTSeparate measurement systems

Different dashboards can depend on the same sensor, crawler, classification model or upstream provider. Interface diversity can conceal instrument dependence.

TEST / ACQUISITION STACK
DIMENSION / TIMESeparate observation events

Repeated captures can show persistence or change. They add temporal evidence, although a stable systematic error may remain shared across every capture.

TEST / EVENT INDEPENDENCE
DIMENSION / CONTROLSeparate decision authority

Common ownership, funding or editorial approval can align what is measured, published or omitted even when individual authors differ.

TEST / GOVERNANCE PATH
DIMENSION / INCENTIVENon-identical pressure to report

Independent organizations may share commercial, political or reputational incentives. Incentive alignment does not invalidate evidence, but it is a possible correlation path.

TEST / ERROR DIRECTION
DIMENSION / INTERPRETATIONSeparate analytical judgments

Several summaries can inherit one expert’s classification or causal interpretation. Diversity at the observation layer may coexist with dependence at the inference layer.

TEST / ANALYST LINEAGE
EVD / 06.13 TRIANGULATION DESIGN

Change the observation path while holding the proposition stable.

Triangulation is deliberate diversification. Each additional line should reduce reliance on a plausible shared failure. Random source accumulation creates volume; designed triangulation creates diagnostic strength.

STRATEGY / SOURCEORIGIN

Independent collectors observe the same bounded state

Use separate origin paths with access to the same subject. Compare collection dates, definitions and scope before treating agreement as convergence. This strategy reduces dependence on one publisher or originating record but may retain a shared method or target-system error.

DIVERSIFIESORIGIN + CONTROL
MAY SHAREMETHOD + TARGET
USE WHENORIGIN BIAS IS PLAUSIBLE
STRATEGY / METHODMECHANISM

Different methods reach the same proposition

Combine direct observation, structured records, documentary evidence and appropriately bounded expert interpretation. Agreement is stronger when no single method failure explains all lines. Confirm that each method actually supports the same claim rather than a neighboring construct.

DIVERSIFIESERROR MODEL
MAY SHAREDEFINITION + SCOPE
USE WHENMETHOD ERROR IS MATERIAL
STRATEGY / TIMEPERSISTENCE

Repeated captures test stability and transition

Observe the same claim across declared windows. Stable agreement supports persistence; change reveals a time-bound state. Temporal triangulation does not remove systematic method bias, so pair it with another method when the measurement system itself is uncertain.

DIVERSIFIESOBSERVATION EVENT
MAY SHAREINSTRUMENT ERROR
USE WHENSTATE VOLATILITY MATTERS
STRATEGY / PERSPECTIVEACCESS

Different positions expose different parts of the system

Operator records, customer observations, regulatory filings and infrastructure telemetry may each reveal a partial state. Convergence across positions can strengthen a shared proposition, while differences may expose boundaries rather than error.

DIVERSIFIESACCESS POSITION
MAY SHAREEVENT + CATEGORY
USE WHENNO SOURCE SEES THE WHOLE
EVD / 06.14 EXTENDED EVIDENCE CASEBOOK

Count the genuinely new information inside each apparent confirmation.

These cases show how the same discipline operates across search, entities, markets, incidents, AI retrieval and digital assets.

CASE / SEARCH RANKINGSERP

Three tools report position four

All three interfaces may rely on separate captures, or two may resell one upstream dataset. Align query, location, language, device and time; then trace acquisition lineage. If independent captures agree within one window, the rank state is corroborated for that environment—not globally or permanently.

CLAIMPOSITION 4 / DECLARED CONTEXT
TESTCAPTURE LINEAGE
LIMITCONTEXT + TIME
CASE / COMPANY IDENTITYENTITY

Registry, official site and trade publication agree

The registry independently supports legal name and incorporation. The official site supports current self-description. A trade article may simply quote that site. The final entity profile should map each field to the evidence line that can observe it rather than treating all three sources as equal support for every attribute.

CLAIMFIELD-SPECIFIC IDENTITY
TESTATTRIBUTE ACCESS
LIMITSELF-REPORTED CLAIMS
CASE / MARKET SHAREMEASURE

Two reports estimate 18 percent

Matching values do not establish independence. Both reports may purchase the same panel and apply similar category boundaries. Inspect provider, panel, denominator and transformation. Independent transaction data with a compatible market definition would add a more diverse path than a third report built on the same panel.

CLAIM18% / DEFINED MARKET
TESTDATASET + DENOMINATOR
LIMITUNOBSERVED TRANSACTIONS
CASE / CYBER INCIDENTEVENT

Status page, users and telemetry report disruption

These perspectives can triangulate occurrence, duration and impact. The status page has operator access, users observe functional consequences and telemetry observes network behavior. Agreement supports a bounded incident state, while severity and root cause require separate evidence.

CLAIMDISRUPTION OCCURRED
TESTTIME + SERVICE ALIGNMENT
LIMITCAUSE NOT YET PROVEN
CASE / BACKLINKREFERENCE

Two indexes discover the same referring page

Independent crawlers can corroborate that a link was observable, provided the canonical source, target, link state and capture period align. A live direct fetch adds method diversity. Agreement does not prove the link was continuously present or that it caused ranking movement.

CLAIMLINK OBSERVED
TESTCANONICAL URL + LIVE FETCH
LIMITNO CAUSAL EFFECT CLAIM
CASE / AI CITATIONRETRIEVAL

Several models cite the same source

Model outputs are separate runs but may share web indexes, training material or ranking signals. Preserve prompt, time, interface and citations. Cross-model agreement can show repeated source selection; it does not independently verify the cited claim unless the underlying source and proposition are validated.

CLAIMSOURCE REPEATEDLY SELECTED
TESTRUN + SOURCE VALIDATION
LIMITSHARED RETRIEVAL UNKNOWN
CASE / PRODUCT FEATURECURRENT STATE

Documentation and live interface agree

Documentation states availability, while a controlled live observation confirms the feature under one plan, region and account type. The methods are different and mutually reinforcing. The conclusion must retain those conditions because availability may differ across tiers or deployments.

CLAIMFEATURE AVAILABLE / BOUNDED
TESTPLAN + REGION + VERSION
LIMITNO UNIVERSAL AVAILABILITY
CASE / CONTENT GAPCOVERAGE

Search results, user tasks and corpus audit converge

A SERP sample shows recurring subtopics, user research identifies unresolved tasks and a page-level audit finds missing evidence units. These methods observe different surfaces. Convergence supports a bounded coverage gap more strongly than keyword repetition alone, while priority still depends on audience and decision objectives.

CLAIMSPECIFIC COVERAGE GAP
TESTINTENT + TASK + CORPUS
LIMITPRIORITY REQUIRES DECISION CONTEXT
EVD / 06.15 CONFIRMATION THRESHOLDS

More evidence is useful only when it changes the justified state.

Set the required degree of corroboration in proportion to consequence, reversibility, volatility and error cost. Low-impact descriptive claims may need modest confirmation; irreversible or safety-critical decisions require stronger, more diverse and more current evidence.

STATE / SINGLE LINEObserved but uncorroborated

One relevant evidence line supports the claim. Report its provenance and limits. Do not imply consensus, but do not discard direct evidence merely because a duplicate has not been found.

USE / DISCOVERY + LOW-COST REVIEW
STATE / REPEATEDSeveral records, one family

The claim is widely repeated but still depends on one origin, dataset or interpretation. This is dissemination evidence, not independent confirmation.

USE / TRACE TO ORIGIN
STATE / CORROBORATEDIndependent compatible evidence

At least two sufficiently independent lines support the same bounded proposition. Shared assumptions and remaining uncertainty stay explicit.

USE / BOUNDED FINDING
STATE / TRIANGULATEDDifferent failure paths converge

Independent origins or methods reach the same proposition through meaningfully diverse observation paths. Confidence rises because one plausible failure is less able to explain all support.

USE / HIGHER-CONSEQUENCE CLAIM
STATE / CONTESTEDIndependent evidence materially disagrees

Agreement is incomplete or a credible contradiction remains. Preserve both sides and continue with conflict diagnosis rather than hiding the minority line.

USE / EVIDENCE CONFLICT REVIEW
STATE / DECAYEDSupport may no longer be current

Past convergence does not guarantee present validity. Volatile states require temporal review, new captures or an explicit historical boundary.

USE / RECOLLECTION OR VERSIONING
EVD / 06.16 CORROBORATION RECORD

Make convergence inspectable after the analysis is finished.

A reusable record explains which evidence lines were counted, why they were considered independent, how the claim was normalized and what limits survived. It allows later evidence to update the conclusion without reconstructing the investigation from memory.

FIELD / CLAIMNormalized proposition

Entity, relation, value, scope, time and qualifiers, plus the original wording from each source.

REQUIRED / YES
FIELD / LINESIndependent evidence families

Origins, methods, timestamps and preserved records, with derivative publications grouped under their upstream line.

REQUIRED / YES
FIELD / DEPENDENCIESShared inputs and failure paths

Common datasets, instruments, ownership, incentives, definitions and target-system dependencies that remain after diversification.

REQUIRED / YES
FIELD / AGREEMENTExact area of convergence

Which parts of the claim agree, permissible tolerance and any dimensions where values remain merely compatible rather than equal.

REQUIRED / YES
FIELD / STATECurrent confirmation classification

Uncorroborated, repeated, corroborated, triangulated, contested or temporally decayed, with a short justification.

REQUIRED / YES
FIELD / REVIEWUnknowns and update trigger

Evidence that would strengthen, weaken, contradict, supersede or require re-collection of the current claim state.

REQUIRED / YES
EVD / 06.17 FREQUENT QUESTIONS

Corroboration and triangulation, without source-count mythology.

Concise answers to the distinctions that most often determine whether apparent agreement adds real evidence.

FAQ / 01MINIMUM

How many sources are required for corroboration?

There is no universal count. Two genuinely independent and claim-fit lines may provide stronger corroboration than twenty derivative articles. The required strength also depends on consequence, reversibility, volatility and the known error of each method.

FAQ / 02PRIMARY

Are primary sources always independent?

No. Two primary reports may share one dataset, instrument, funding structure or classification method. “Primary” describes proximity to origin; independence describes whether the error paths are sufficiently separate.

FAQ / 03AGREEMENT

Does convergence prove a claim is true?

No. Convergence raises justified confidence when independent, compatible paths agree, but all paths may still share an unknown assumption or observe the wrong construct. State the remaining common dependencies.

FAQ / 04METHOD

What makes method triangulation strong?

The methods must have different relevant failure modes while supporting the same normalized proposition. Using several interfaces backed by one acquisition system is not meaningful method diversity.

FAQ / 05TIME

Can repeated captures corroborate a changing state?

They can corroborate persistence within an interval or reveal a transition. Each capture remains time-bound. Repetition across time does not support an unqualified claim that the state is permanent.

FAQ / 06CONFLICT

What if independent sources disagree?

Do not average or vote automatically. Align identity, definition, scope, time, unit, method and lineage. If material disagreement remains, classify the claim as contested and move to explicit conflict resolution.

FAQ / 07AI

Do several AI answers count as independent sources?

Not by default. Models may share training material, retrieval indexes, cited sources or ranking signals. Treat outputs as separate run records, trace their cited evidence and avoid inferring independence from different interfaces alone.

FAQ / 08REPORT

What should a corroboration conclusion say?

Name the bounded claim, number of visible records, number of independent evidence families, dimensions diversified, shared dependencies, disagreement state, confidence and the event that would trigger review.

EVD / 06.18 CLAIM-SPECIFIC CORROBORATION PLANS

The claim type determines what independent support must look like.

Corroboration is strongest when collection is designed around the failure modes of the proposition. These plans show why one generic “find three sources” rule cannot protect every type of conclusion.

PLAN / IDENTITYWHO OR WHAT

Legal identity and public identity require different evidence

For legal name, incorporation and registered office, prefer an authoritative registry record plus an independently preserved version where possible. For current public identity, compare the official site, current product interface and independent references. Map parent, subsidiary, brand and predecessor relations rather than merging them. Stop when each required attribute has a claim-fit evidence line and no unresolved entity collision remains.

PRIMARY RISKENTITY CONFLATION
DIVERSIFYREGISTRY + LIVE PRESENCE
STOP RULEATTRIBUTE-LEVEL SUPPORT
PLAN / CURRENT STATESTATUS

Volatile states need independent captures and a short validity window

A current price, ranking, feature, policy or availability claim can expire quickly. Use at least one direct capture with complete environment details and a second independent observation when consequence justifies it. Preserve time, location, device, account state and version. Agreement supports the state only within the declared window. Re-collect when volatility or decision delay makes the evidence stale.

PRIMARY RISKTEMPORAL DECAY
DIVERSIFYCAPTURE EVENT + METHOD
STOP RULEVALIDITY WINDOW SATISFIED
PLAN / NUMERIC ESTIMATEMAGNITUDE

Matching numbers are weak when denominator and model are shared

For search demand, traffic, market share, audience or valuation estimates, reconstruct construct, population, period, unit, denominator and transformation. Seek a second dataset or measurement model with a meaningfully different error path. Report compatible ranges when exact values cannot be observed directly. Stop when uncertainty is narrow enough for the intended decision—not when several dashboards display similar precision.

PRIMARY RISKSHARED MODEL + FALSE PRECISION
DIVERSIFYDATASET + ESTIMATION METHOD
STOP RULEDECISION-RELEVANT RANGE
PLAN / CAUSAL CLAIMWHY

Repeated association does not independently establish causation

Several observational reports can share the same confounding structure. Corroborate causal claims through designs that test alternative explanations: controlled comparison, natural experiment, temporal ordering, mechanism evidence or negative controls. Align the intervention and outcome definitions. Stop only when plausible rival explanations have been reduced to a level proportionate to the consequence of acting on the claim.

PRIMARY RISKSHARED CONFOUNDING
DIVERSIFYDESIGN + MECHANISM
STOP RULERIVALS EXPLICITLY TESTED
PLAN / HISTORICAL EVENTPAST STATE

Contemporaneous records and later interpretation play different roles

Use records produced near the event to establish what was documented at the time, while recognizing access and incentive limits. Later independent research can triangulate context, identity and consequence but may inherit earlier archives. Separate primary record, witness account, institutional record and retrospective analysis. Stop when the bounded event claim is supported across appropriate paths and material contradictions are disclosed.

PRIMARY RISKRETROSPECTIVE COLLAPSE
DIVERSIFYCONTEMPORARY + RETROSPECTIVE
STOP RULEEVENT BOUNDARY STABLE
PLAN / AI OUTPUTGENERATED STATE

Corroborate the run separately from the factual content

Repeated outputs can establish that several controlled runs produced a pattern, such as citation frequency or answer stability. They cannot establish that the generated factual claims are true. Preserve prompt, model or interface, time, location, session context and cited sources. Then validate each material claim against independent evidence. Stop rules must distinguish output reproducibility from factual corroboration.

PRIMARY RISKOUTPUT-TRUTH CONFUSION
DIVERSIFYRUNS + SOURCE VALIDATION
STOP RULETWO EVIDENCE LAYERS REPORTED
PLAN / NEGATIVE CLAIMABSENCE

Failure to observe is not automatically evidence of absence

To support “no record,” “not available” or “does not occur,” define where the item would be expected to appear, how thoroughly that space was observed and the detection limits. Diversify search paths, indexes, time windows or access positions. A bounded absence claim may be justified even when universal absence is not. Stop when coverage and detection sensitivity match the stated boundary.

PRIMARY RISKINCOMPLETE OBSERVATION
DIVERSIFYSEARCH PATH + COVERAGE
STOP RULEDETECTION LIMIT DECLARED
PLAN / FORECASTFUTURE

Independent forecasts may still share assumptions

Compare data lineage, model family, scenario assumptions and forecast horizon. Several forecasts based on the same growth rate or macroeconomic premise form a dependent cluster. Use structurally different models and scenario stress tests, then track calibration against outcomes. Stop when the range and downside are decision-ready; do not turn cross-model agreement into certainty about an unobserved future.

PRIMARY RISKASSUMPTION CORRELATION
DIVERSIFYMODEL FAMILY + SCENARIO
STOP RULEBOUNDED FORECAST RANGE
EVD / 06.19 FINAL QUALITY CONTROL

Before declaring corroboration, attempt to break it.

A short adversarial review tests whether the confirmation survives lineage, scope and method challenges. Passing these controls does not create certainty; it shows that the stated confidence is proportionate to the inspected evidence.

CONTROL / ECHOCollapse every derivative lineage

Recalculate support after mirrors, quotations, press-release copies and shared datasets are grouped. If the conclusion loses most of its support, report repetition rather than corroboration.

QUESTION / HOW MANY ORIGINS REMAIN?
CONTROL / SCOPENarrow every claim to its evidence

Remove unsupported geography, populations, periods and universal language. The conclusion should remain useful even after its real boundary becomes visible.

QUESTION / WHERE DOES SUPPORT END?
CONTROL / FAILUREName one error that could fool every line

Look for common definitions, target-system behavior, incentives, acquisition infrastructure and hidden assumptions. If one plausible failure explains all agreement, triangulation is incomplete.

QUESTION / WHAT IS STILL SHARED?
CONTROL / CONTRADICTIONSearch for the strongest contrary line

Do not test only whether support exists. Identify credible disagreement and explain whether it reflects scope, time, method or a material unresolved conflict.

QUESTION / WHAT WOULD WEAKEN THE CLAIM?
CONTROL / CONSEQUENCEMatch confirmation depth to error cost

A reversible low-impact choice and an irreversible high-impact commitment should not share the same evidence threshold. Increase diversity and currency as consequence rises.

QUESTION / WHAT HAPPENS IF WRONG?
CONTROL / UPDATEDeclare the reopening condition

Name the new record, observed change, failed prediction or elapsed validity window that would trigger reassessment. A conclusion without an update rule becomes resistant to evidence.

QUESTION / WHEN MUST WE LOOK AGAIN?
Release corroboration only when the evidence count, independent-line count and dependency map are all visible.

The public conclusion should state the exact proposition, applicable boundary, degree of convergence, surviving shared assumptions and current review condition. This makes agreement useful without converting it into certainty and lets future evidence strengthen, narrow or contest the claim without erasing its history.

VISIBLERecords found
VISIBLEIndependent evidence families
VISIBLEShared dependencies and limits
EVD / CONTINUE CONNECTED EVIDENCE TOPICS

Corroboration maps convergence. Conflict analysis owns divergence.

EVD/06 establishes whether support is independent and multi-modal. EVD/07 diagnoses material disagreement through scope, timing, lineage, method and source quality.

EVD / PRINCIPLE 06 · TOPICALAUTHORITY.ORG

Count paths, not echoes. Triangulate what can fail.

Resolve lineages, collapse dependencies, compare compatible observations and diversify source, method or time. Report convergence without deleting uncertainty.

EVIDENCE / EVD-06 · CORROBORATION & TRIANGULATIONCLAIM → LINEAGE → INDEPENDENCE → CONVERGENCE → QUALIFICATION
TAO / CONTACT · DIRECT TRANSMISSION Have an asset, domain or market position to investigate? ENTER CONTACT SYSTEM →
DIGITAL ASSET INTELLIGENCE + EXECUTION
EXECUTED BY
BB DIGITALNA AGENCIJA

Investigation, consulting and execution of digital assets, premium-domain strategies, information architecture, semantic systems, websites and agreed digital growth plans.

TOPICALAUTHORITY.ORG / SEMANTIC INTELLIGENCE SYSTEM BB DIGITALNA AGENCIJA / BB.HR