TOPICALAUTHORITY.ORG TAO / ROOT

What Is Decision Intelligence?

TOPICALAUTHORITY.ORG / DIGITAL ANALYSISFOUNDATION CONCEPTS / METHODS / LIMITS
ANL / 01 · FIRST-PRINCIPLES DEFINITION

What Is DigitalAnalysis?

Digital analysis is not the act of looking at data. It is a declared, inspectable transformation of bounded digital observations into patterns, explanations and findings that remain traceable to evidence.

ANL.OS / ANALYTICAL CONTRACTVALID STRUCTURE
ANL
QUESTION / DIFFERENCEWhere does demand exceed relevant supply?
01OBJECTDefined query territory
02UNITNormalized query–intent group
03BOUNDARYLanguage, market, device and time
04COMPARATORRelevant existing result supply
05FINDINGBounded, testable demand gap
INPUTEVIDENCE
OPERATIONINTERPRET
OUTPUTFINDING
STATE / 01Object
STATE / 02Observation
STATE / 03Transformation
STATE / 04Comparison
STATE / 05Interpretation
STATE / 06Finding
ANL / 101 DEFINITION

Analysis is a controlled change of state.Inputs become findings through visible operations.

A dataset does not contain its own explanation. The analyst defines the object, unit of analysis, boundary, transformation rules and comparison frame before assigning meaning to a pattern.

Digital analysis is the systematic transformation and comparison of bounded observations to produce a tested interpretation or decision-relevant finding.

The word digital describes the observed environment, not a weaker standard of reasoning. Pages, queries, links, entities, transactions, API records, logs and market signals remain incomplete representations of a changing system. Their meaning depends on identity resolution, collection conditions, time, missingness and the relationship between the measure and the concept it is intended to represent.

INPUTTraceable observations and derived measures
OPERATIONDeclared normalization, segmentation and comparison
INTERPRETATIONMeaning tested against alternatives
LIMITScope, uncertainty and unresolved contradiction
OUTPUTA bounded finding, not an unrestricted truth claim
ANL / 102 ANALYTICAL GRAMMAR

A finding has dependencies.Write them into the analytical grammar.

The notation below is not a universal formula or scoring model. It exposes the minimum dependency structure: the finding changes when observations, boundaries, transformations, comparators or assumptions change.

ANL-G01 / DEPENDENCY NOTATION

No transformation, no analytical claim.

Collection establishes what was available to observe. Analysis begins when a declared operation changes the representation of those observations so that structure, difference, relationship or change can be tested.

F = Interpret [ Compare ( T(OB), C ), A ]
OBObservations inside boundary B
TDeclared transformations
CComparator or baseline
AExplicit assumptions
FBounded finding
URetained uncertainty
DEPENDENCY RULE: if B, T, C or A changes materially, F must be recalculated and may no longer describe the same analytical question.
ANL / 103 TERM BOUNDARIES

Analysis is not a synonymfor every research activity.

Select each term. The distinction is functional: several activities can occur in one project, but each produces a different output and requires different controls.

OPERATION / INTERPRET

Analysis

Transforms and compares observations to identify structure, difference, relationship or change, then tests the resulting interpretation.

PRIMARY OUTPUTA bounded analytical finding with visible dependencies and uncertainty.
NOT SUFFICIENTA chart, summary or dashboard without a declared analytical question.
ANL / 104 EPISTEMIC STATES

Preserve what each statement is.Never flatten the states.

The same sentence can move through several states during analysis. Labeling prevents an analyst’s interpretation from being presented later as if it were directly returned by the source.

01Observed

A value, event or expression directly captured within the declared boundary.

SOURCE-BOUND
02Derived

A value calculated, normalized, grouped or classified from observations.

RULE-BOUND
03Patterned

A repeated structure or difference detected across comparable units.

MODEL-BOUND
04Interpreted

A proposed meaning tested against alternatives, context and contradictions.

ASSUMPTION-BOUND
05Found

The narrowest conclusion supported after testing and uncertainty control.

EVIDENCE-BOUND
ANL / 105 ANALYSIS COMPILER

The analytical object determineswhat the result can mean.

Switch the lens. Each analytical object requires a different unit, comparator and valid output. Reusing one metric across all four lenses produces category errors, not efficiency.

ANL-C01 / LENS COMPILERASSET LENS
OBJECT / IDENTIFIABLE ASSETCOMPILED

What position can this asset hold?

Resolve the asset before measuring its performance. Identity, control, distinctiveness, addressable territory and existing associations define what can legitimately be compared.

UNITOne resolved asset across declared surfaces
COMPARATORFunctionally relevant assets in the same territory
VALID OUTPUTPosition, differentiation and addressable opportunity
INVALID LEAPA strong name guarantees market demand or authority
ANL / 106 WORKED EXAMPLES

Two objects. Two valid analyses.One discipline of separation.

These illustrative records demonstrate reasoning structure, not live market findings. Observations, derivations, interpretations and limits remain visibly separated.

EXAMPLE / ASSETILLUSTRATIVE

Category domain position

Question: can one exact-match domain become a useful reference center within a bounded service category?

OBSERVEDDomain identity, current pages, indexed queries, referring sources and existing market entities are preserved.
DERIVEDQuery territory and competitor supply are normalized by intent, geography and service equivalence.
INTERPRETEDThe name is distinctive and structurally aligned with one category, but recognition remains unproven.
LIMITDomain quality alone does not create demand, authority, trust or execution capability.
FINDINGThe asset is a plausible category container worth testing through bounded coverage and repeated observation.
NEXT TEST: publish the declared territory, observe discovery and compare branded and non-branded association over time.
EXAMPLE / DEMANDILLUSTRATIVE

Unresolved query territory

Question: does a group of queries indicate one recurring need that existing result supply does not satisfy well?

OBSERVEDQuery forms, result types, ranking pages, recurring modifiers and visible user tasks are collected.
DERIVEDLexical variants are grouped only where intent, expected outcome and entity relationship align.
INTERPRETEDRepeated task language suggests coherent demand while current supply remains fragmented.
LIMITLow supply may reflect low value, access constraints or measurement blind spots rather than opportunity.
FINDINGA bounded demand gap exists only if the need recurs and the proposed asset can resolve it better.
NEXT TEST: collect adjacent queries, inspect result satisfaction and test whether one structured answer improves retrieval and task completion.
ANL / 107 EXTENDED CASE FILES

One analytical grammar.Eight radically different objects.

The following cases show how the same discipline changes with the object under inspection. Each case separates what was captured, what was calculated, what can be interpreted and what remains unsupported. The records are illustrative, but the reasoning structure is operational.

CASE / DOMAIN ASSETIDENTITY BEFORE PERFORMANCE

Can a category domain support a defensible market position?

The object is not merely a string of characters. It includes ownership continuity, canonical use, recognizable meaning, competing interpretations, existing associations and the territory in which the name is expected to function. The question is therefore narrower than whether the domain “sounds strong.”

OBSERVEDThe exact domain, registration continuity, live use, indexed pages, branded references, query associations and similarly named market entities are captured at a declared date.
DERIVEDReferences are classified by source type, query intent and entity identity. Ambiguous mentions are excluded rather than assigned to the asset by lexical similarity alone.
COMPAREThe asset is compared with names serving the same function and market, not with every domain containing the same word. Category alignment, recall, distinctiveness and addressable coverage are treated as separate dimensions.
LIMITA descriptive or memorable name does not establish trust, commercial demand, authority, conversion or execution quality. Those are distinct objects requiring their own observations.
FINDINGThe domain can be described as a structurally suitable category container when its identity is clear and its semantic territory is coherent. Market recognition remains a hypothesis until repeated external evidence appears.
NEXT TEST: publish bounded category coverage, observe independent association and compare recognition across synchronized periods.
CASE / SEARCH ENVIRONMENTRESULT COMPOSITION

Has a search result environment changed?

A ranking list is a time-bound surface generated under specific language, location, device and query conditions. Analysis begins only after those conditions are held stable enough to compare. A different result at a later date may reflect ordinary rotation, personalization, collection error or a structural change.

OBSERVEDOrdered results, result types, visible domains, titles, features, cited sources and capture conditions are preserved for every collection window.
DERIVEDURLs are resolved to canonical pages, pages to domains, and result features to stable classes. Duplicate surfaces and collection failures are marked rather than counted as meaningful absence.
COMPAREResult overlap, domain concentration, page-type share and positional persistence are measured across equivalent windows. A single moved URL is separated from system-level composition.
LIMITThe observed result surface does not expose the complete ranking mechanism. Movement cannot be attributed to one cause merely because a site change occurred beforehand.
FINDINGA structural result change is supportable when multiple composition measures move together and the new state persists beyond expected rotation. The finding concerns the observed environment, not an undisclosed mechanism.
NEXT TEST: repeat synchronized captures, isolate update boundaries and check whether the changed composition survives query reformulation.
CASE / DEMAND TERRITORYNEED BEFORE VOLUME

Do many query forms represent one demand?

Lexical similarity is not sufficient for aggregation. Two phrases may share most words while expecting different actions, entities or outcomes. Conversely, different vocabulary may express the same unresolved task. Demand analysis therefore requires a controlled grouping rule before totals are calculated.

OBSERVEDQueries, modifiers, result composition, recurring tasks, destination pages, seasonality and available demand measures are collected within one market and time boundary.
DERIVEDVariants are grouped only when intent, expected resolution and entity relationship align. Informational discovery, evaluation and transaction are preserved as different states even when they concern the same topic.
COMPAREThe normalized need is compared with existing supply, adjacent need clusters and prior windows. Raw volume, concentration and unmet-task evidence remain separate indicators.
LIMITLarge query volume does not equal commercial value, and weak supply does not automatically indicate an opportunity. The need may be low-value, inaccessible, seasonal or already resolved outside search.
FINDINGA coherent demand territory exists when multiple expressions repeatedly converge on the same task and expected outcome. An addressable gap exists only if a proposed asset can satisfy that task more completely.
NEXT TEST: test cluster stability under new query samples and inspect whether one answer architecture resolves the shared task.
CASE / ENTITY SYSTEMRESOLUTION UNDER AMBIGUITY

Are scattered references describing the same entity?

An entity is not established by a matching label alone. Names collide, organizations change, products inherit brand language and locations reuse terms. Analysis must combine identifiers, attributes, relations and temporal context before references can be merged.

OBSERVEDNames, aliases, official identifiers, addresses, ownership statements, linked profiles, dates, products and source provenance are preserved without premature consolidation.
DERIVEDCandidate records are linked through explicit matching rules. Exact identifiers carry more weight than shared descriptions; contradictory dates or locations lower confidence and remain visible.
COMPAREEach candidate is tested against alternative identities and against the minimum attribute set expected for the claimed entity class. Missing evidence is not silently replaced by similarity.
LIMITA high similarity score is not identity proof. It may describe a branch, former name, reseller, unrelated organization or copied profile.
FINDINGReferences may be consolidated when independent identifiers and relations converge without material contradiction. Otherwise the correct analytical output is an unresolved identity set with explicit competing candidates.
NEXT TEST: seek one independent identifier capable of separating the highest-probability candidates.
CASE / CONTENT SYSTEMCOVERAGE AS STRUCTURE

Does a large corpus actually cover its subject?

Page count is an inventory measure, not a coverage conclusion. A corpus can contain thousands of pages while repeating one narrow template, missing central entities or leaving important relationships unexplained. Coverage analysis must define the subject universe and distinguish presence from adequate treatment.

OBSERVEDCanonical pages, headings, entities, claims, internal links, query associations, update dates and duplicate patterns are collected for the bounded corpus.
DERIVEDPages are mapped to concepts, intents and relationships. Repeated mentions are separated from substantive explanations, and orphaned coverage is distinguished from integrated coverage.
COMPAREThe observed map is compared with a declared subject model containing required concepts, dependencies, edge cases and user tasks. Depth and connectivity are assessed separately.
LIMITCompleteness against one model does not establish universal completeness or ranking authority. The model itself may omit emerging concepts or privilege one interpretation of the field.
FINDINGThe corpus has strong structural coverage when important concepts are substantively treated, connected through meaningful paths and aligned with distinct user tasks. Remaining gaps must be stated by type, not reduced to one score.
NEXT TEST: challenge the subject model with external terminology, unresolved queries and expert counterexamples.
CASE / LINK GRAPHRELATION BEFORE COUNT

Does the linking structure support navigation and meaning?

A link count says little without source, destination, anchor, context and graph position. Ten links from duplicated navigation do not perform the same function as one contextual edge connecting a prerequisite to a dependent explanation.

OBSERVEDSource URL, destination URL, anchor, surrounding text, placement, follow state, canonical status and crawl accessibility are preserved for each edge.
DERIVEDEdges are classified as navigational, hierarchical, contextual or referential. Template repetition is normalized, redirects are resolved and broken destinations remain explicit failures.
COMPAREObserved paths are compared with the intended dependency graph and major user journeys. Reachability, depth, concentration and semantic fit are inspected together.
LIMITCentrality does not equal importance in every context, and dense linking does not guarantee clarity. Excessive edges can erase hierarchy and increase routing ambiguity.
FINDINGThe graph supports the system when important nodes are reachable through semantically justified paths, hierarchy remains legible and no essential branch depends on one fragile edge.
NEXT TEST: remove dominant template edges and verify whether contextual structure still exposes the intended relationships.
CASE / EXTERNAL SIGNALREFERENCE QUALITY

Do external references constitute meaningful support?

External references differ in independence, context, editorial control, persistence and relationship to the claim under inspection. Treating every mention as equal converts a heterogeneous evidence set into a misleading total.

OBSERVEDReferring source, exact destination, surrounding statement, publication date, ownership relation, link state and later availability are preserved.
DERIVEDSources are classified by independence and function. Syndicated copies, network-owned properties and mechanically generated pages are grouped so repetition is not mistaken for corroboration.
COMPAREThe reference profile is compared by relevant source class, contextual fit, concentration and temporal persistence rather than by total count alone.
LIMITA reference can indicate awareness without endorsing quality, and a link can exist without transferring trust, users or topical meaning.
FINDINGExternal support is stronger when independent, relevant sources repeatedly connect the asset to the same bounded subject through explicit context. The conclusion must name that subject and cannot expand beyond it.
NEXT TEST: inspect whether the same association persists across independent source classes and later observation windows.
CASE / RETRIEVAL SYSTEMSELECTION BEFORE CITATION

Why is one source repeatedly retrieved?

Visibility inside an answer system is the result of several separable events: eligibility, retrieval, selection, synthesis and citation. A cited page proves that one output referenced it under one condition; it does not reveal the entire retrieval process or guarantee repeated selection.

OBSERVEDPrompt, locale, model surface, response, cited sources, citation placement, answer claims and collection time are captured together.
DERIVEDRepeated prompts are grouped by task rather than wording alone. Sources are resolved, claims are mapped to citations and uncited answer segments remain distinct.
COMPARESelection frequency, source diversity, claim alignment and persistence are compared across controlled prompt variants and time windows.
LIMITCitation frequency is not a universal authority score. Outputs can vary by model, retrieval index, prompt wording, availability and product changes.
FINDINGA source has stable retrieval visibility only when it is repeatedly selected for the same task across reasonable variations. The supported finding is conditional on the tested system, prompts and period.
NEXT TEST: hold the task constant, vary expression and measure whether source selection survives.
ANL / 108 FAILURE-MODE MATRIX

Every analytical shortcutcreates a predictable distortion.

The matrix names failures by the layer where they enter. Detection matters because a polished chart can remain internally consistent while answering the wrong question, comparing incompatible units or overstating what the evidence permits.

FAILURE / 01Object substitution

The measured object silently changes from page to domain, query to intent, mention to entity or traffic to demand. The numbers may be accurate while the conclusion describes something else. Prevention requires naming the unit in every table and checking whether aggregation changes its identity.

CONTROL / FREEZE THE UNIT
FAILURE / 02Boundary drift

Language, geography, device, date range or inclusion rules change between observations. Apparent movement may therefore be produced by a different research universe. Prevention requires versioned boundaries and a new series whenever comparability is materially broken.

CONTROL / VERSION THE SCOPE
FAILURE / 03Proxy inflation

A convenient metric is treated as if it directly measured authority, quality, trust, demand or readiness. Prevention requires an operational definition explaining which part of the concept the proxy represents and which parts remain unobserved.

CONTROL / NAME THE DISTANCE
FAILURE / 04Invalid denominator

A percentage changes because the available universe changed, not because the numerator improved. Indexed share, coverage rate and citation rate are meaningless when eligible totals are undefined or unstable. Prevention requires publishing numerator, denominator and missing units together.

CONTROL / EXPOSE THE BASE
FAILURE / 05Convenient comparator

The analyst selects rivals, periods or baselines that make the focal object appear stronger. Prevention requires a functional inclusion rule established before results are inspected and sensitivity tests using plausible alternative comparison sets.

CONTROL / PREDECLARE THE SET
FAILURE / 06Missingness blindness

Absent observations are interpreted as zero, nonexistence or failure even though collection may be incomplete. Prevention requires an explicit missing state, acquisition diagnostics and a test of whether missingness is concentrated in one class or period.

CONTROL / PRESERVE UNKNOWN
FAILURE / 07Aggregation collapse

A mean or total hides opposite movements across segments. Overall visibility can rise while every priority market falls if the composition of the measured set changes. Prevention requires distribution checks and stratified results before a global summary is accepted.

CONTROL / INSPECT THE PARTS
FAILURE / 08Temporal leakage

Information available after an event is allowed to influence classification of an earlier state. This creates unrealistically clean explanations and invalid predictive claims. Prevention requires observation cutoffs and reconstruction from information available at the declared time.

CONTROL / LOCK THE CLOCK
FAILURE / 09Correlation narration

Two measures move together and are converted into a causal story without mechanism, timing or alternative explanations. Prevention requires causal ordering, rival hypotheses, intervention evidence and explicit refusal to infer cause when those conditions are absent.

CONTROL / TEST ALTERNATIVES
FAILURE / 10Metric optimization

Once a visible measure becomes the target, behavior adapts to improve the measure without improving the underlying condition. Prevention requires counter-metrics, outcome checks and periodic review of whether the operational definition still represents the intended concept.

CONTROL / WATCH THE PROXY
FAILURE / 11False precision

Decimals, composite scores or confidence labels imply more certainty than sampling, model stability or source quality supports. Prevention requires uncertainty intervals, coarse categories where appropriate and disclosure of sensitivity to reasonable methodological choices.

CONTROL / MATCH RESOLUTION
FAILURE / 12Finding expansion

A local, historical or platform-specific result is written as a universal present-tense truth. Prevention requires every conclusion to retain its object, boundary, period, evidence class and strongest unresolved limitation.

CONTROL / BOUND THE CLAIM
ANL / 109 VALIDITY CONTROLS

A useful result must survive inspection.Six controls keep it analytical.

Visual polish, computational complexity and dataset size cannot compensate for an unresolved object, invalid comparison or conclusion that exceeds the evidence. These controls operate together: passing five does not cancel a material failure in the sixth.

CONTROL / 01Object validity

The analyzed object is uniquely resolved and stable enough to compare. Its identity, granularity and relationships are explicit. If the object changes during aggregation, the output is relabeled and interpreted at the new level.

ASK / WHAT EXACTLY?
CONTROL / 02Measure validity

The selected measure represents the intended concept rather than a convenient proxy. Known distance between indicator and concept is stated, while unsupported dimensions remain outside the conclusion.

ASK / MEASURES WHAT?
CONTROL / 03Boundary stability

Locale, language, device, time and inclusion rules remain consistent. Any material boundary revision creates a new analytical state and triggers recollection, renormalization or an explicit comparability break.

ASK / INSIDE WHICH SYSTEM?
CONTROL / 04Comparison validity

Units and baselines are functionally comparable after declared normalization. The comparator is selected by a reproducible rule, not because it produces the preferred contrast.

ASK / COMPARED TO WHAT?
CONTROL / 05Interpretive restraint

The proposed explanation is tested against plausible alternatives, contradictions and sensitivity to analytical choices. Description, relationship and cause remain different claim classes.

ASK / WHAT ELSE FITS?
CONTROL / 06Finding proportionality

The conclusion is no broader, stronger or more current than its evidence. It identifies what would falsify or revise it and preserves material unknowns instead of hiding them inside a score.

ASK / HOW FAR CAN IT GO?
ANL / 110 SENSITIVITY TESTS

A result earns weightby surviving reasonable alternatives.

Sensitivity analysis changes one defensible decision at a time and observes whether the finding remains materially intact. It does not search endlessly for a version that destroys the result. The alternatives must be plausible, declared and connected to a genuine source of analytical uncertainty.

TEST / BOUNDARYINCLUSION STABILITY

Would a reasonable boundary change reverse the result?

Suppose a content audit reports 78 percent conceptual coverage. That value depends on which concepts enter the reference universe, how variants are consolidated and which pages qualify as substantive treatment. A strong result should not collapse merely because a small number of defensible edge concepts are added.

BASEThe primary model contains 120 required concepts defined before the corpus is evaluated. Ninety-four meet the declared coverage threshold.
ALTERNATIVEA broader model adds eighteen adjacent concepts supported by external terminology. A narrower model removes twelve specialist concepts irrelevant to the intended audience.
OBSERVECoverage becomes 71 percent under the broad model and 82 percent under the narrow model. The direction remains strong, but the exact percentage is model-sensitive.
DISCLOSEThe valid finding is not “the subject is 78 percent complete.” It is that coverage is consistently substantial across three defensible boundaries, with weakness concentrated in the same two subtopics.
DECISION RULE: preserve the range and stable gap locations; do not promote one model-dependent point estimate as objective completeness.
TEST / TRANSFORMATIONNORMALIZATION STABILITY

Does the pattern exist only under one calculation?

A demand comparison can change when raw totals, population-adjusted rates, share of category or logarithmic transformations are used. None is universally correct. Each answers a different question, and the analytical task is to identify whether the central finding depends on one fragile representation.

BASEMarket A has twice the raw query count of Market B. The initial result suggests materially greater demand.
ALTERNATIVECounts are normalized by addressable population, total category activity and active business entities. Brand-driven queries are also removed in a separate specification.
OBSERVEMarket A remains larger in absolute terms, while Market B has higher per-capita intensity and a greater non-branded share. The original statement survives only for total observable scale.
DISCLOSEThe analysis must separate market size from demand intensity. Combining them into one ranking would conceal two different strategic conditions.
DECISION RULE: select the transformation that matches the question and report divergent views when they represent distinct valid constructs.
TEST / COMPARATORREFERENCE-SET STABILITY

Is the advantage real or manufactured by the rival set?

An asset may look exceptional against a broad collection of weak or irrelevant sites and ordinary against specialists solving the same task. Comparator design must follow functional equivalence: audience, market, purpose and operating constraints should be sufficiently aligned for the difference to carry meaning.

BASEThe focal asset has greater topic coverage than twenty general market websites selected by keyword overlap.
ALTERNATIVEThe set is restricted to eight specialist assets serving the same audience and expanded to include four adjacent substitutes competing for the same user task.
OBSERVEThe asset remains broader than substitutes but falls behind specialists in evidence depth, update recency and entity precision. Its advantage is architectural breadth, not universal quality.
DISCLOSEA single winner label would destroy the analytical result. Different comparator sets expose different competitive dimensions and should remain visible.
DECISION RULE: state the functional set, preserve dimension-level outcomes and test whether the claimed advantage survives plausible membership changes.
TEST / TIMEWINDOW STABILITY

Does the conclusion survive another valid time window?

Digital systems change through updates, seasonality, publication cycles, migrations and market events. A start date selected after seeing the graph can manufacture acceleration, decline or stability. Time sensitivity therefore tests rolling windows, event boundaries and measurement continuity.

BASEA twelve-week series shows rising discovery of newly published pages and appears to support a durable improvement.
ALTERNATIVEThe analysis is repeated across eight-, twelve- and sixteen-week windows, then recalculated after excluding a migration week and separating newly eligible pages.
OBSERVEThe direction remains positive in longer windows but weakens substantially after eligibility growth is controlled. Faster discovery explains part, not all, of the observed increase.
DISCLOSEThe evidence supports improved discovery rate within the measured period. It does not establish permanent performance, ranking growth or the cause of the remaining change.
DECISION RULE: retain the shortest stable claim shared by defensible windows and mark structural breaks instead of smoothing them away.
ANL / 111 OPERATING PROCEDURE

Run analysis as an inspectable procedure.Eight steps, no invisible leaps.

The order can iterate, but the dependencies remain. A changed boundary requires recollection or renormalization; a contradicted interpretation requires a revised finding.

01Declare the question

Specify whether the analysis seeks structure, difference, relationship, change or explanation.

QUESTION
02Resolve the object

Define the exact asset, demand set, entity, system or temporal process.

IDENTITY
03Set the boundary

Freeze inclusion, exclusion, geography, language, channel and time.

SCOPE
04Preserve observations

Retain provenance, missingness and acquisition context before transformation.

EVIDENCE
05Transform explicitly

Document normalization, classification, aggregation and derived measures.

RULES
06Compare valid units

Apply a relevant baseline, cohort, rival set, prior state or counterfactual.

DELTA
07Test interpretation

Inspect contradictions, alternatives, sensitivity and potential error.

CONTROL
08Bound the finding

State what follows, what does not and which next observation can change it.

OUTPUT
ANL / 112 FINDING CONTRACT

A finding is not a slogan.It is a bounded analytical object.

The final statement must remain connected to the question, observed units, transformation rules, comparator and uncertainty that produced it. Removing those dependencies may make the sentence shorter, but it also changes its meaning and often inflates its certainty.

Within [declared boundary], object X differs from comparator Y on measure M after transformation T; the result survives tests S, remains sensitive to condition K and does not establish unsupported claim Z.

A complete finding does more than report a number. It identifies the object and period, states the direction and magnitude where appropriate, names the comparison frame and exposes the condition under which the interpretation weakens. It also distinguishes the result from the action someone may later choose. For example, evidence that one demand territory is underserved may justify further testing, but it does not by itself authorize investment, predict revenue or prove that one organization can execute the opportunity.

The strongest useful finding is not the strongest sentence that can be written. It is the most decision-relevant statement that remains true across defensible analytical choices. If changing one reasonable threshold reverses the conclusion, threshold sensitivity is part of the finding. If identity remains unresolved, the result belongs to a candidate set rather than one named entity. If the measured period ends before a major event, the conclusion remains historical and cannot be silently updated into the present.

OBJECTName the exact unit or resolved system the finding describes.
CONDITIONRetain geography, language, channel, device, population and period.
OPERATIONIdentify the transformation and comparison responsible for the result.
STABILITYState which alternative models, windows or thresholds were tested.
UNCERTAINTYPreserve missing observations, ambiguity and plausible competing explanations.
NON-CLAIMExplicitly name the causal, predictive or universal leap not supported.
ANL / 113 RESEARCH STACK

Analysis has one exact position.It cannot replace the layers around it.

The research stack separates acquisition, procedural control, claim support, interpretation and action. Collapsing these functions makes the final output impossible to inspect.

LAYER / LABSProduce observations

Instruments retrieve, calculate or expose defined digital states and measurements.

OUTPUT / OBSERVATION
LAYER / METHODSGovern procedure

Questions, scope, sampling, transformations and reproducibility control how observations are produced.

OUTPUT / PROCEDURE
LAYER / EVIDENCEAuthorize support

Provenance, relevance, integrity and corroboration determine what a claim can rely on.

OUTPUT / SUPPORT RELATION
LAYER / ANALYSISInterpret meaning

Declared transformations and comparisons produce tested, bounded findings from evidence.

OUTPUT / FINDING
LAYER / APPLICATIONInform action

A decision, strategy or intervention uses the finding under its own constraints and accountability.

OUTPUT / ACTION
CONTINUE DIGITAL ANALYSIS

Choose the analytical methodthat matches the question.

Continue from the foundational definition into descriptive, exploratory, comparative, relational, temporal and causal analysis. Each guide addresses a distinct analytical task, the evidence it requires and the conclusions it can legitimately support.

TOPICALAUTHORITY.ORG / DIGITAL ANALYSISFOUNDATIONAL GUIDE / METHODS AND LIMITS
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