The Three Worlds of Analytics

Why Business Meaning, Data, and Material Data Need Different Governance

Version 1.1

Huayin Wang
DOI: 10.5281/zenodo.22146487
Publication date: 28 August 2026

Analytics is often described as a path from a business question to data and from data to an answer. That description compresses three different worlds into one.

The Business World is where customers, stores, products, organizations, policies, classifications, domain relationships, and concepts such as active customer or recognized revenue acquire organizational meaning.

The Data World is where measures, measure families, anchors, universes, lineage, reducers, derivations, support, absence, consistency, and data identity exist under data law.

The Material Data World is where numbers, strings, blobs, records, rows, columns, dataframes, tables, schemas, files, databases, partitions, indexes, and execution plans are represented, stored, changed, moved, and computed.

These are not merely three layers of one software stack. They differ in their objects and ontology; processes and operations; technologies; languages and codes; governance rules and instruments; clients and expertise; and jurisdiction.

Analytics is unusual because ordinary analytical work repeatedly crosses all three. A business question must become a data object. That object must be realized through material data and computation. The result must return as a data result and often cross again into business interpretation, claims, decisions, and action.

The crossings happen all the time. The authority on each side is not the same.

Three worlds, three jurisdictions

The Business World

The Business World is the world of domain meaning. Its central question is: What does this mean in this organization or domain?

An organization can decide what counts as an active customer. Finance can define recognized revenue. Operations can determine what qualifies as an open store. Its operations include defining, naming, classifying, relating, interpreting, mapping, approving, and revising business concepts. Its technologies include business ontologies, semantic layers, catalogs, metric systems, business-rule systems, and domain models. Its languages include ontology languages, semantic-model declarations, metric definitions, taxonomies, controlled vocabularies, and business rules. Its governance instruments include ownership, domain authority, definition review, taxonomy, policy, approval, stewardship, and semantic consistency.

Business-world technologies are often hybrid in practice. A semantic model may contain joins, aggregation declarations, material bindings, or other computational instructions. That does not erase the jurisdictional distinction: carrying a computation declaration does not by itself confer authority over data identity or data law.

Business authority is authority over business meaning. It is not automatically authority over data law or material representation.

The Data World

The Data World contains measures, families, anchors, universes, data identities, lineage, conceptual mappings among data objects, reducers, derivations, movement, support, and absence. Its central question is: What data object exists, and what follows lawfully from it?

If the business defines an open store, that definition does not by itself establish whether an average over open stores can be answered when one open store has no revenue observation. If Finance defines revenue, that authority does not determine whether a movement duplicates the measure, whether a reducer may lawfully be applied again, or whether two objects carrying the same display name have the same data identity.

Its operations include constitution, anchoring, mapping, reduction, derivation, movement, validation, support evaluation, and consistency checking. Its governance instruments include identity law, family and reducer law, movement law, universe and existence law, support, derivability, and consistency. Its clients include analysts, statisticians, analytical engineers, analytical services, and intelligent systems.

The Material Data World

The Material Data World is the world of representation, storage, mutation, movement, and computation. Its objects include numbers, strings, booleans, timestamps, blobs, records, rows, columns, arrays, dataframes, relations, tables, schemas, files, databases, partitions, indexes, and execution plans.

Its central question is: How is data represented, stored, changed, retrieved, and computed?

Its processes include create, read, insert, update, delete, populate, copy, extract, load, transform, ETL and ELT, filter, join, aggregate, sort, materialize, index, partition, persist, replicate, query, and execute. Its technologies include databases, warehouses, lakehouses, dataframe systems, query engines, storage engines, ETL tools, orchestration systems, and computational infrastructure. Its languages include SQL, DDL and DML, relational algebra, dataframe APIs, ETL specifications, database APIs, engine configuration, and execution plans. Its governance instruments include schemas, data types, keys, constraints, transactions, permissions, storage policies, orchestration controls, data-quality checks, execution correctness, reliability, and security.

Material presence is not data identity. A row may be present in a table without constituting the data object a question requires. A row may be absent without establishing that the data value is zero.

The differences are structural

Dimension	Business World	Data World	Material Data World
Objects	customers, products, stores, contracts, policies, KPIs, domain relationships	measures, families, anchors, universes, lineage, support, data identities	bits, bytes, numbers, strings, blobs, records, rows, columns, arrays, dataframes, tables, schemas, files, streams, databases
Work	define, classify, relate, interpret, approve	constitute, anchor, map, reduce, derive, move, validate	create, read, update, delete, populate, ETL, filter, join, aggregate, materialize
Technologies	ontologies, semantic layers, catalogs, metric systems	data constitutions, data languages, validation systems	databases, warehouses, dataframe systems, query engines, ETL tools
Languages	ontology languages, semantic models, metric definitions, business rules	governed data expressions, lineage and derivation representations	SQL, DDL/DML, dataframe APIs, ETL definitions, execution plans
Governance	ownership, definitions, domain authority, taxonomy, policy	identity, reducer law, movement law, existence, support, derivability, consistency	schemas, types, constraints, transactions, permissions, material-data quality, execution correctness
Expertise and clients	business users, domain experts, semantic modelers	analysts, statisticians, analytical engineers, analytical services	applications, data engineers, DBAs, platform engineers, compute engines

Scope: worlds are jurisdictions, not the three foundations

The three worlds are not another way of dividing Data · Certainty · Intelligence, and they are not stages of an analytical service.

They answer a narrower question:

What kinds of things are being governed, and what kind of authority applies to each?

Questions of evidence, certainty, claims, inference, interpretation, and standing can arise as analytics moves within and across these worlds. Evidence is not a fourth world, and Intelligence is not a layer above the diagram.

A small vocabulary for the Data World

The Data World does not require a reader to learn an entire formal calculus before the jurisdiction can be recognized. Five ordinary ideas are enough to see the distinction:

• Identity --- what data object is this, and what makes it the same object across representations?
• Population / universe --- what set of things does the object range over?
• Reduction --- what operations may combine values, and under what conditions?
• Support / absence --- what observations or grounds are present, missing, zero, or ineligible?
• Lawful derivation --- what other data objects may legitimately be produced from this one?

The Theory of Data develops a richer interior vocabulary --- including families, anchors, lineage, movement, and reducer classes --- but those terms refine the jurisdiction rather than create it.

Why the middle is the Data World

A measure may have governed identity and law before any table stores its values.

That simple fact separates the Data World from the Material Data World.

The word data is pulled in two directions. Business and semantic systems pull it toward the world the data describes. Database and computational systems pull it toward the material forms that store and carry it.

The Data World is neither of those things.

A measure can have a governed identity, anchor, family, lineage, reducer law, and derivation even when no table currently stores its values. The same data object may later be realized as a dataframe, a database relation, a Parquet file, an Arrow array, an API response, or an in-memory value without becoming a different data object merely because its material form changed.

The converse also holds. A stored number, row, or column can exist without establishing which governed data object it realizes, at what anchor, over what universe, with what lineage, or under which laws.

It helps to separate three ideas that ordinary speech often compresses into exists:

• a constituted data object has governed identity even if no values are currently stored;
• a derivable data object can be lawfully produced from other governed data;
• a materialized result has actually been computed or stored in some material form.

So when we say that data can exist without materialization, we mean that the data object can already have identity and law. We do not mean that an uncomputed value has somehow already been observed.

Two facts can therefore hold at once:

Data can exist without being materialized.

Material data can exist without establishing data identity.

Material data carries, stores, moves, and computes data. It does not exhaust what data is.

The distinction is not simply syntax versus semantics. A table or dataframe can carry rich meaning and still be only one materialization of a data object. The deeper question is whether changing the material form changes what data object it is.

Data is neither the world it describes nor the materialization that carries it.

Business semantics can declare computation without governing data law

The distinction between the Business World and the Data World does not depend on pretending that semantic layers contain only names and definitions.

Modern semantic systems can declare aggregation functions, metric formulas, and join relationships. A semantic model may specify SUM(amount) or define how a metric reaches a table.

Those declarations are important. They are still different from governing the laws of the resulting data object.

Declaring SUM(amount) as a computation instruction does not by itself establish the reducer’s closure class, whether another reduction is admissible, whether movement across a relationship preserves the measure, or whether two derivation paths are consistent.

Likewise, declaring a join relationship does not by itself establish the data consequences of movement across that relationship.

The distinction is therefore not semantics versus computation.

It is business/domain authority versus data-law authority.

The causal order

The Data World does not exist because an architecture inserts it between business meaning and material systems.

It is already there.

The Theory of Data describes its ontology and governing laws. Analytics then has a structural fact to deal with: useful work must cross between this independently existing Data World and its two neighboring worlds.

That causal order matters:

Data World → theory → governance problem → architecture

not:

architecture → invented middle layer.

Any architecture may take advantage of this structure. No particular architecture owns the distinction.

Three worlds imply two governed interfaces

The three worlds imply two crossings: Business World ⇄ Data World and Data World ⇄ Material Data World. These are interfaces between worlds, not additional worlds of their own.

This also helps locate two familiar Analytical Governance problems without turning the three worlds into a service pipeline.

The intent gap lies primarily over the upper crossing:

Does the representation in the Data World correctly capture the user’s intent from the Business World?

The servability gap lies primarily over the lower crossing:

Can the Material Data World provide what the Data World request requires?

Analytics crosses all three

Consider: What was average revenue per open store yesterday?

Open store begins as business meaning. The requested average belongs to the Data World: it has a population, anchor, measure identity, reducer, and support requirements. The observations used to realize it belong to the Material Data World.

An ordinary analytical path therefore looks like:

business meaning → data constitution → material realization → data result → business claim or action.

Intermingling hides the crossings

For much of analytics history, people carried these crossings implicitly. That allowed familiar substitutions: business meaning for metric definition; metric definition for SQL; schema for data ontology; material row presence for data existence; missing row for zero; query success for validity of the data result; business name for data identity; material-data quality for Data World support.

Each substitution can work in some circumstances. None is a general law. The problem begins when one world silently answers a question that belongs to another.

The 47, 48, and 50 problem

Suppose 50 stores exist. 48 were open yesterday. Revenue observations arrived for 47.

Someone asks:

What was average revenue per open store yesterday?

The Business World can establish what open store means and which stores qualify.

The Material Data World can establish which revenue observations are present.

A database can correctly sum the 47 observations and divide by 47.

But the Data World request asks for an average over the governed population of open stores.

What happened to the forty-eighth store? Was its revenue zero? Was the observation missing? Did the feed fail?

The material observations cannot settle those questions by themselves. Nor can the business definition of open store.

The point here is jurisdictional: three kinds of authority establish different facts. Business authority establishes what open means. Material-data authority establishes what observations are present. Data authority determines what data object those facts jointly establish.

Governance becomes confused when authority is confused

Modern organizations often place business definitions, semantic ownership, metric governance, schemas, lineage, access control, data quality, analytical validation, transformation policy, and AI permissions under the broad heading of data governance. These are all legitimate concerns. They do not govern the same kinds of objects.

When governance treats the three worlds as one, it becomes difficult to answer: Who is authoritative about what?

When jurisdiction becomes an ownership conflict

The distinction becomes concrete when something changes.

Suppose Finance changes the approved meaning of recognized revenue. That is a legitimate Business World change. The new meaning may require data objects to be reconstituted. But Finance’s authority over the definition does not silently authorize a new reducer law or make previously invalid derivations valid.

Now suppose a platform team changes a column from nullable to non-nullable, adds a schema constraint, or migrates a table. Those are legitimate Material Data World changes. They may improve material integrity. They do not, by themselves, establish the governed population of a measure or prove that data support is complete.

So the question who is authoritative? has no single organizational answer. Authority follows the kind of object being governed and the question being asked about it.

Intelligent agents raise the stakes

Intelligent agents can now interpret business intent, resolve semantic concepts, choose analytical operations, find data, generate queries, execute tools, interpret results, and recommend actions. One probabilistic component can cross all three worlds in seconds.

Those are multiple crossings between different kinds of authority, not one decision. The more capable the agent becomes, the more important it becomes to know where one authority ends and another begins.

Closing

The three worlds need one another. They are still different worlds.

Analytics binds them together. That is its power. It is also one of its deepest governance challenges.

The task is not to eliminate the crossings.

It is to stop treating a crossing as though nothing changed when we crossed it.