What Kind of Thing Is This?

Before we can ask whether something is true, important, useful, dangerous, beautiful, fair or causal, we need to ask a quieter question:

A person is not the same kind of thing as a statement about a person. A measurement is not the same kind of thing as the property being measured. A photograph is not the event it depicts. A law is not the behaviour it regulates. A prediction is not the future. A map is not the territory. A category is not an individual member of the category. A model is not the world.

These distinctions sound obvious when stated slowly. In real reasoning, however, they collapse into one another surprisingly easily. Human conversations do it. News reports do it. Databases do it. Institutions do it. Search engines do it. Artificial-intelligence systems do it. Once unlike things are treated as if they were interchangeable, errors propagate.

This article develops a practical grammar for keeping them apart.

Featured Snippet: The Root Definition

The hidden grammar of reality is the set of distinctions that lets us identify what kind of thing we are dealing with before we reason about it. Useful kinds include entities, agents, properties, states, events, processes, relationships, quantities, observations, claims, records, rules, systems, models, signals, questions and unknowns.
The central rule is simple: do not confuse the thing with information about the thing.

The Simple Answer

Reality contains different kinds of things, and our representations of reality contain different kinds of things too.

A student is an entity and an agent.
“The student is tired” is a claim about a state.
A thermometer reading is an observation or measurement record.
“The room is 29°C” is a claim formed from a measurement.
A school timetable is an artifact that represents intended events.
A lesson is an event or process unfolding through time.
A school rule is a norm.
The school itself is an organisation and system.
A diagram of the school is a model.
A rumour about the school is a signal carrying an uncertain claim.
“We do not know why attendance fell” is not a fact about absence; it is a fact about our current knowledge.

Good reasoning begins when these types stop being blurred together.

Why This Question Matters

Suppose someone says:

“This school is excellent.”

What kind of thing is that sentence?

It is not the school. It is not excellence itself. It is not necessarily an observation. It is a claim, probably an evaluative one. To examine it, we need to ask what evidence supports it, what definition of “excellent” is being used, whose viewpoint matters, what time period is meant, what comparison set applies and whether the claim is about examination results, student well-being, teaching quality, facilities, value, culture or something else.

The sentence becomes more useful when its type is clear.

This is the first major principle of the article:

Classification should often happen before evaluation.

Before asking whether a thing is good, bad, true, false or causal, ask what kind of thing it is.

Two Worlds Must Be Kept Apart

A useful first division is between the world and our representations of the world.

1. Things that occur or exist

people
animals
organisations
objects
places
properties
states
events
processes
relationships
systems
physical quantities

2. Things that represent, describe or regulate what occurs or exists

words
names
records
photographs
maps
measurements
claims
reports
models
predictions
rules
plans
questions
labels
database rows

The distinction is not always perfectly sharp. A printed map is also a physical object. A law is both a document and a social rule. A person can be represented by a database record while also being the author of another record. The point is not to force the world into one box. The point is to preserve enough type information that reasoning remains honest.

Type 1: Entity

An entity is something we treat as having identity. It can be referred to as the same thing across at least some interval of time or across multiple statements.

Examples include a person, a school, a river, a book, a company, a building, a country, a molecule, a planet or a particular computer.

Entity questions include:

Which particular thing is this?
What identifies it?
When did it begin and end?
Has its name changed?
Is this the same entity as the one mentioned elsewhere?
Is this an individual entity or a class of entities?

Identity matters because names are unreliable. Two people may share a name. One institution may have several names. A company may rebrand without becoming a wholly new organisation. A river may cross political boundaries. A school may move location while remaining institutionally continuous.

Name is not identity.

Type 2: Agent

An agent is an entity that can act, decide, respond or participate in action in a way that matters to the question being asked.

People are agents. Groups can be treated as agents in some contexts. Governments, companies, committees and software systems can also function as agents when they select actions according to goals, procedures or control structures.

But agency should not be attributed casually. “The market wants lower prices” may be a shorthand description of many interacting buyers and sellers, not evidence that a single market-agent exists with a mind and intention.

The useful questions are:

Who or what acted?
What capacity did it have?
Was the action intentional, automatic or emergent?
Was the apparent agent actually a collection of agents?
Who had authority?
Who had information?
Who bears responsibility?

Agent is therefore not merely a synonym for entity. An entity can exist without being the acting unit relevant to the case.

Type 3: Property

A property is an attribute we use to describe an entity, event, process or other thing.

Examples include mass, colour, temperature, age, enrolment size, location, speed, price, legal status and ownership.

A property is not automatically a measurement. Temperature can be a physical property; “29°C” can be a measured value; “the thermometer showed 29°C” is an observation record; and “the room was 29°C at 2 pm” is a claim based on that record.

One piece of reality has already become four different information types.

Type 4: State

A state describes how something is at a particular time or during an interval.

A door can be open. A student can be absent. A battery can be charged. A road can be closed. A company can be insolvent. A machine can be idle.

States are useful because they make time explicit. “The road is closed” is incomplete if the closure ended yesterday. “The student is weak in algebra” may be misleading if it compresses a temporary performance state into a permanent identity.

A good state description therefore asks:

State of what?
At what time?
According to what observation?
How certain are we?
Is the state persistent or temporary?
What would count as a state change?

Type 5: Event

An event is something that happens. It is bounded in time strongly enough that we can refer to its occurrence.

Examples include an election, an examination, a collision, a graduation, a launch, a court decision, an earthquake, a purchase, a goal in a football match or a meeting.

An event can contain many smaller events and can itself form part of a larger process. A school examination has a start, a sequence of activities and an end. A national election includes nominations, voting, counting and certification. Event boundaries depend partly on the scale of analysis.

This introduces a crucial idea:

The kind of thing something is can depend on granularity.

At one scale, a lesson is an event. At another, it is a process containing explanation, questioning, practice, feedback and revision.

Type 6: Process

A process is organised change through time.

Learning is a process. Digestion is a process. Manufacturing is a process. Erosion is a process. Publishing is a process. Investigation is a process. Recovery is a process. Budgeting is a process.

Processes differ from isolated events because their internal sequence matters. If a process is interrupted, reordered or deprived of one stage, the outcome may change.

Useful process questions include:

What are the stages?
What enters the process?
What leaves it?
What transforms?
Where are the decision points?
What can fail?
What feedback loops exist?
What is reversible?
What is irreversible?

Calling a process a “thing” without recognising its temporal structure is a common source of bad explanation.

Type 7: Relationship

A relationship connects two or more things.

Examples include parent-of, located-in, owns, employs, teaches, causes, follows, competes-with, part-of, member-of, resembles, contradicts and depends-on.

Relationships are easy to flatten into properties, but the distinction matters. “Alice is a teacher” can describe a role. “Alice teaches Ben” describes a relationship. “School A owns Building B” links two entities. “Event X occurred before Event Y” links two events through temporal order.

Relations also have direction. Parent-of is not the same as child-of. Owns is not the same as owned-by. Causes is not the same as caused-by.

A good information system should preserve that direction rather than storing a vague notion that two things are merely “connected”.

Type 8: Quantity and Value

A quantity is something that can be expressed with magnitude. A value is a particular assignment to a quantity under specified conditions.

Numbers are powerful because they feel precise. Precision, however, is not the same as correctness.

“120” is almost meaningless by itself.

120 what?
Measured where?
Measured when?
By what method?
With what units?
With what uncertainty?
Is it a count, rate, score, percentage, index or measurement?

Numbers need type, unit, scope and provenance.

“Rainfall was 120 mm in the observation period” is a claim about a measured quantity. “120 mm” is the value. The gauge reading is a record. The rainfall itself is a physical process. The weather station is an entity. The measuring procedure is a method. The reliability of the reading is an epistemic question.

Again, one sentence can contain many different kinds of thing.

Type 9: Observation

An observation is an encounter with evidence: something detected, measured, seen, heard, recorded or otherwise registered by a person or instrument.

Observation does not guarantee truth. A witness can misperceive. A sensor can malfunction. A test can be contaminated. A questionnaire can be biased. A camera can show only one angle.

But observation is still different from claim.

“The sensor logged 42.3” is a statement about an observation record. “The machine was overheating” is an interpretation that may be supported by that observation. “The machine failed because it overheated” is a causal claim requiring more evidence.

This gives us an important ladder:

world
observation
record
interpretation
claim
explanation
decision

Errors often arise when several rungs are skipped.

Type 10: Record and Artifact

A record is an artifact intended to preserve information. An artifact is something produced or shaped for a purpose.

Examples include a report, photograph, transcript, database row, certificate, receipt, map, chart, letter, contract, textbook, spreadsheet, court filing or log file.

Records are essential because human memory is limited and events disappear into the past. But records are not transparent windows onto reality.

Every record has provenance:

Who created it?
When?
For what purpose?
From which source?
Under what procedure?
Has it been altered?
What does it omit?
What version is this?

A receipt records a transaction; it is not the transaction itself. A photograph records light from a scene; it is not the whole scene. A certificate records an institutional judgement; it is not the person’s complete ability. A database record represents an entity; it is not the entity.

The record-reality distinction is one of the most important disciplines in information work.

Type 11: Claim

A claim is a proposition presented as being the case.

Claims can be true, false, partly true, outdated, ambiguous, context-dependent, unfalsifiable or currently unresolved.

Examples:

“Water boils at 100°C at standard atmospheric pressure.”
“The meeting starts at 9 am.”
“This treatment reduced symptoms in the study group.”
“The policy caused the decline.”
“This is the best school.”

These are not equally structured. The first is a scientific statement with specified conditions. The second is a schedule claim that can become outdated. The third is an empirical study claim. The fourth is causal. The fifth is evaluative and underspecified.

Therefore a fact-checking system should not merely store “claim = sentence”. It should ask what kind of claim it is.

descriptive
quantitative
historical
causal
comparative
predictive
evaluative
normative
definitional
attributive

Different claim types require different evidence.

Type 12: Rule, Norm and Requirement

A rule or norm says what should, must, may or may not happen within a system.

Rules include laws, examination regulations, classroom expectations, software permissions, building codes, contractual terms and game rules.

A rule is not the same thing as a regularity.

“People usually queue here” describes a regular pattern. “People must queue here” states a rule. “Please queue here” is an instruction. “Queueing is fair” is an evaluative claim.

Blurring these types produces confusion between what happens, what is allowed, what is expected and what is considered desirable.

Type 13: System

A system is a set of interacting parts whose relationships matter to the behaviour of the whole.

Schools, ecosystems, transport networks, electrical grids, financial markets, families, computers, organisations and supply chains can all be studied as systems.

System questions differ from object questions.

What are the parts?
How are they connected?
What flows through the system?
What are the boundaries?
What enters and leaves?
What feedback exists?
Where are the bottlenecks?
What happens if one component fails?
Which behaviours emerge from interaction rather than from a single part?

A system can also contain entities, states, events, processes, rules and records at the same time. System is therefore a higher-order organisational type.

Type 14: Model

A model is a selective representation built to help us understand, predict, explain, design or decide.

A globe is a model of Earth. A diagram is a model. A mathematical equation can be a model. A timetable models planned activity. A financial forecast models possible future performance. A curriculum map models an intended learning progression.

Models are useful because they leave things out.

That is also why models can mislead.

The right question is not only “Is the model correct?” but:

Correct enough for what purpose?
At what scale?
Under what assumptions?
For which time period?
Which variables are omitted?
Where does it stop working?

A model is not a failed copy of reality. It is a purposeful compression of reality.

Type 15: Signal

A signal is something that carries information or potentially informative variation.

Signals can be words, sounds, sensor outputs, gestures, prices, warning lights, traffic patterns, search trends or changes in behaviour.

Signal is not meaning.

A fire alarm is a signal. Its intended meaning is that a possible fire condition requires attention or evacuation. The actual presence of fire is a separate state of the world. A false alarm proves the distinction.

Similarly, a rising price is a signal produced by a market process. It may reflect scarcity, demand, speculation, supply disruption, policy, expectations or some combination. The signal does not contain its own explanation.

Type 16: Question, Unknown and Void

Not knowing is itself information.

A question identifies a gap we want to close. An unknown is a value, explanation, identity or state that has not been established. A void is a recognised absence in the available information or structure.

This is different from saying the answer is zero, false, absent or impossible.

Consider:

“No rainfall occurred.”
“Rainfall was not measured.”
“The measurement is missing.”
“The measurement was zero.”
“We do not know whether rainfall occurred.”

These are five different states of knowledge.

Data systems often collapse them into one blank cell. Reasoning systems should not.

The Great Separation: Thing, Evidence, Claim and Decision

Many reasoning failures can be prevented by preserving four layers:

Thing: What exists or happened?
Evidence: What was observed or recorded?
Claim: What is being asserted from the evidence?
Decision: What action should be taken?

For example:

A student answers 8 of 20 algebra questions correctly.

The answer sheet is a record. The marked responses are observations or derived records. “The student scored 40% on this test” is a quantitative claim. “The student has not yet mastered these algebra skills” is an interpretation. “The student is bad at mathematics” is a much broader claim and may be unjustified. “Give targeted practice on factorisation” is a decision.

Keeping these layers distinct makes education more humane and more accurate. A performance state is not a permanent identity.

Category Is Not Instance

A category is a class used to group things. An instance is a particular member.

“Dog” is a category. A particular dog called Milo is an instance. “Secondary school” is a category. A particular school is an instance. “Volcano” is a category. Mount Etna is an instance.

Categories are useful compressions. They let us carry knowledge from known examples to new ones. But categories also create danger: we may assume every instance has every typical feature.

The correct question is:

Which properties follow from category membership, and which merely happen to be common?

This matters in science, education, law, medicine, history and ordinary life.

Role Is Not Identity

A person can be a parent, teacher, passenger, customer, voter, patient, witness and neighbour at different times. These are roles, not complete identities.

Organisations also occupy roles. A university may be an employer in one relationship, a research institution in another and a property owner in a third.

Role-based reasoning is powerful because it avoids attaching temporary functions permanently to entities.

Event Is Not Explanation

“A happened, then B happened” describes temporal order.

It does not by itself prove that A caused B.

Cause is a relationship claim requiring an appropriate evidential basis. Depending on the domain, that basis may involve controlled experiments, quasi-experiments, mechanistic evidence, statistical analysis, historical documents, comparative cases or other methods.

The grammatical error is simple:

sequence was mistaken for causation.

Measurement Is Not the Measured Thing

Modern institutions are full of measurements: marks, rankings, prices, rates, counts, indices, temperatures, speeds, test results and survey scores.

Measurements are representations produced through methods. The method matters.

A student’s examination score is evidence about performance under particular test conditions. It is not the student. GDP is a measure of economic activity defined by a statistical framework. It is not “the economy” in its entirety. A blood-pressure reading is a measurement. It is not the whole state of a patient’s cardiovascular health.

Measurement is valuable because it compresses complex reality into comparable form. The compression must not be forgotten.

Record Is Not Memory of the World

Archives, libraries and databases preserve traces. They do not preserve everything.

What survives is shaped by selection, preservation, format, institutional practice, power, accident and loss. A surviving record can be authentic yet incomplete. A complete file can still represent only one institutional viewpoint. Multiple independent records may strengthen confidence, but agreement does not automatically remove shared bias.

This is why provenance is not an administrative nuisance. Provenance is part of meaning.

Prediction Is Not Future Fact

A prediction is a claim about a possible future state or event. Even a highly accurate predictive system produces forecasts, not future facts.

Probability language should therefore remain attached to predictions. A 70% forecast is not a weak statement pretending to be certain. It is a different kind of statement.

Once the event occurs, we can compare the prediction with the outcome. Until then, the prediction remains epistemically distinct from the future it describes.

Plan Is Not Outcome

A plan is an artifact representing intended action. A schedule represents intended timing. A budget represents intended allocation. A policy represents a chosen rule or direction. None of these guarantees execution.

Therefore:

planned ≠ completed
announced ≠ implemented
funded ≠ delivered
scheduled ≠ occurred
approved ≠ effective

Institutional reasoning becomes much cleaner when intention, execution and outcome are represented separately.

Definition Is Not Discovery

Some disagreements are about facts. Others are about definitions.

If two people use the same word differently, gathering more evidence may not resolve the disagreement until the term is clarified.

For example, “success”, “intelligence”, “wealth”, “quality”, “fairness” and “civilisation” can be defined in different ways depending on purpose and discipline.

A definition is a rule for how a term is being used. It may be conventional, technical, legal, operational or stipulative. Definitions can be more or less useful, coherent or appropriate, but they are not empirical observations in the same way that a temperature reading is.

Absence Is Not Unknown

This distinction deserves repetition because it is one of the most common failures in data reasoning.

If a database does not contain a record of an event, several possibilities exist:

the event did not occur;
the event occurred but was not recorded;
the record exists elsewhere;
the record was lost;
the search failed;
access is restricted;
the naming convention differs;
the relevant time period was not searched.

“No record found” is therefore not automatically equivalent to “did not happen”.

That small grammatical discipline prevents enormous historical, legal and analytical mistakes.

Boundary Is a Kind of Thing Too

Many disagreements are actually disagreements about where a system, event, category or responsibility begins and ends.

When does a recession begin? Where does a neighbourhood end? Which emissions belong to a company’s activity? When does a learning task become an assessment? Which components belong to a transport system?

Boundaries may be physical, legal, administrative, conceptual or analytical.

A boundary is therefore not always “out there” waiting to be discovered. Sometimes it is a decision needed to make analysis possible. Good reasoning states the boundary instead of hiding it.

Time Is Part of Type

Many facts are only facts within a time interval.

A person holds an office for a period. A shop is open during specified hours. A law applies from one date until amendment or repeal. A student belongs to a class during a school year. A price is valid at a time. A web page contains particular content in a particular version.

Therefore a robust claim often needs a temporal field:

valid from
valid until
observed at
published at
effective date
event date
retrieved at

“Who is the minister?” is not a timeless question. “What was the policy?” may require knowing which version and which date.

Time turns apparently simple nouns into versioned realities.

Place and Jurisdiction Change Meaning

A rule, qualification, price, weather observation, school system, legal status or cultural convention may be valid in one place and not another.

Location is therefore not a decorative field. It can determine meaning.

For laws and regulations, jurisdiction is especially important. “Legal” is incomplete without asking legal where, when and under which authority.

Perspective Changes the Claim Type

“The lesson was difficult” may describe a student’s experience. “The lesson contained advanced material” is a content claim. “Most students scored poorly afterwards” is a quantitative observation. “The lesson was badly taught” is an evaluative and causal interpretation.

These are related but different.

Perspective does not make everything subjective. It tells us which layer of reality a statement belongs to.

The Hidden Grammar as a Fact-Checking Pipeline

A practical fact-checking workflow can begin with type before truth.

Step 1: Identify the expression

What exact sentence, number, image, record or assertion are we examining?

Step 2: Identify the referent

What in the world is the expression supposed to refer to?

Step 3: Assign candidate types

Is it an entity, state, event, process, observation, claim, record, rule, model, prediction or something else?

Step 4: Add scope

time
place
jurisdiction
population
measurement unit
version
granularity

Step 5: Identify provenance

Where did the information come from? Primary record? Secondary report? Witness? Sensor? Database? Publication? Anonymous repost?

Step 6: Match evidence to claim type

A historical claim, causal claim, legal claim, scientific claim and evaluative claim do not require identical evidence.

Step 7: Separate confidence from truth value

We may believe a claim is likely true without having conclusive evidence. We may also discover that a once well-supported claim has become outdated.

Step 8: Preserve unresolved states

If evidence is insufficient, record “unknown”, “disputed”, “not yet verified” or another suitable epistemic state. Do not force a binary answer merely because a database field prefers one.

A Compact Machine-Readable Grammar

For computational reasoning, a useful public-facing structure might represent information through fields such as:

type — entity, event, state, process, claim, record, rule, model, observation, unknown
identity — which particular thing
role — what function the thing occupies in this context
time — when it exists, applies or was observed
place — where
jurisdiction — under what authority
relation — how it connects to other things
source — where the information came from
evidence — what supports it
confidence — how strongly it is supported
version — which revision
status — active, historical, planned, disputed, superseded, unknown

The exact vocabulary can vary. The discipline matters more than the labels.

Why This Helps Artificial Intelligence

Artificial intelligence frequently receives language in which several logical types are compressed into a single sentence.

Consider:

“The company announced a new factory that will create 2,000 jobs.”

This can contain:

a company entity;
an announcement event;
a factory artifact or planned facility;
a future-oriented plan;
a quantitative prediction about jobs;
a claim attributed to a source;
a future state that has not yet occurred.

If the system collapses everything into “fact”, it may later report the 2,000 jobs as if they already exist. If it preserves type, it can say: the company announced a plan; the job figure is projected; subsequent evidence is required to establish realised employment.

That is a large improvement in truthfulness created by a small improvement in grammar.

Why This Helps Search

Search improves when queries are typed.

“Mars” may seek an entity. “Mars temperature” seeks a property or measured range. “Mars missions” seeks events and projects. “Mars colonisation” may seek plans, predictions and arguments. “Who discovered Mars?” is partly a question-definition problem because Mars was visible to ancient observers and was not “discovered” in the same sense as a previously unknown object.

A typed query can therefore guide both retrieval and verification.

Why This Helps Education

Students often struggle not because they lack facts but because they confuse logical types.

They give an example when asked for a definition.
They give a description when asked for an explanation.
They give correlation when asked for cause.
They give evidence when asked for a conclusion.
They give a conclusion without evidence.
They confuse a formula with the phenomenon it models.
They treat a diagram as literal reality.
They state a rule when asked what actually happened.

Teaching students to ask “what kind of answer is required?” is therefore closely related to teaching them to ask “what kind of thing is this?”

Why This Helps Science

Science depends on disciplined movement between observation, measurement, model, hypothesis, prediction and explanation.

A model can fit observations and still be replaced by a better model. An observation can be genuine while its interpretation changes. A measurement can be precise while systematically biased. A hypothesis can be plausible while not yet well tested.

Scientific reasoning is powerful partly because it maintains these distinctions instead of treating every statement as the same kind of knowledge.

Why This Helps History

History deals with events that are no longer directly observable. We work through surviving records, artifacts, testimony, later accounts and interpretations.

A letter written in 1910 is an artifact from 1910. Its content may include claims about earlier events. A historian writing in 2026 may interpret the letter. A database may catalogue the letter. A later article may quote the database entry instead of the original document.

Each step creates a new information object with its own provenance.

Historical reasoning becomes stronger when the chain is visible.

Why This Helps News and Media Literacy

A headline can describe an event, summarise a report, quote a claim, repeat a prediction or offer an interpretation. Readers benefit from identifying which one they are seeing.

“Scientists warn X may happen” is not the same as “X happened”.

“Government announces plan” is not the same as “plan completed”.

“Study finds association” is not the same as “A causes B”.

“Witness says” is not the same as “investigation establishes”.

These are grammatical distinctions before they are political or ideological ones.

Why This Helps Archives and Libraries

Libraries and archives do more than store information. They preserve type, context and provenance.

A manuscript, edition, translation, scan, catalogue record and quotation are related but different objects. If those distinctions collapse, authority becomes difficult to trace.

The same is true in digital systems. A cached copy, original publication, revised page, database export and machine-generated summary should not silently become one undifferentiated “source”.