Knowledge is not one kind of thing.
Some knowledge is factual. Some is conceptual. Some is procedural. Some is embodied in models. Some is carried as claims with evidence. Some is uncertain, provisional or disputed. Some is local to one domain or context, while other knowledge transfers widely.
Categorising knowledge well means preserving these differences without pretending that every useful idea fits one permanent hierarchy.
Quick answer: how should knowledge be categorised?
- Facts: statements about what is the case.
- Concepts: organised meanings used to group and distinguish things.
- Models: simplified representations of systems or mechanisms.
- Procedures: knowledge of how to do something.
- Rules: explicit conditions or constraints.
- Claims: propositions that may require evidence or remain disputed.
- Explanations: accounts of why or how something happens.
- Skills: capabilities that may not be fully captured by text.
- Uncertainty: what is unknown, provisional, disputed or probabilistic.
- Provenance: where the knowledge came from and how it was established.
This article applies the architecture from How to Categorise Anything directly to knowledge systems.
1. Define the knowledge unit
A sentence, concept, theorem, method, lesson, model and whole discipline are different units. Classification should preserve level.
2. Facts are not concepts
“Water boils at a given temperature under defined conditions” is a factual claim; “boiling” is a concept used to organise many observations.
3. Concepts create reusable structure
Concepts let many facts become intelligible as members of a larger pattern.
4. Definitions are knowledge artefacts too
A definition states the intended boundary of a concept under a given system or discipline.
5. Rules differ from descriptive facts
Rules may describe logic, law, procedure or convention. Their authority matters.
6. Procedures encode how
A procedure organises steps, conditions, decisions and sequence rather than merely describing a state of affairs.
7. Skills are not fully reducible to procedures
Knowing the written steps for swimming, teaching or surgery is not identical to being able to perform them competently.
8. Tacit knowledge deserves recognition
Some expertise is embodied in pattern recognition, timing and judgement that practitioners cannot easily articulate completely.
9. Models are structured simplifications
A model preserves some aspects of reality and deliberately omits others.
10. Models should be classified by purpose
Explanatory, predictive, descriptive, simulation and decision models serve different jobs.
11. A model is not the thing itself
Knowledge architecture should distinguish entity, representation and claim about the entity.
12. Explanations organise causes and mechanisms
An explanation connects observations, models and causal claims into an account of how or why something occurs.
13. Predictions are future-oriented claims
Predictions should record time horizon, assumptions, method and uncertainty.
14. Claims need evidence relationships
A claim may be supported, challenged, qualified or superseded by evidence. Store those relationships explicitly.
15. Evidence is not the same as claim
An observation can support several claims, and one claim can depend on many pieces of evidence.
16. Hypotheses are provisional claims
A hypothesis is structured enough to test while remaining open to rejection or revision.
17. Theories are organised explanatory systems
Theory should not be used as a synonym for casual guess. Domain context and evidence status matter.
18. Laws and principles need authority and scope
A scientific law, legal rule and design principle are different knowledge types even when all are called laws or principles informally.
19. Examples are knowledge supports
Examples help establish recognition and transfer but should not be mistaken for definitions.
20. Counterexamples are boundary knowledge
A counterexample can reveal where a concept, rule or model stops working.
21. Misconceptions are structured error knowledge
Knowing common wrong models can be highly useful in education and diagnosis because it predicts how reasoning may fail.
22. Knowledge has domain scope
A principle valid in one discipline or jurisdiction may not transfer unchanged to another.
23. Context can change applicability
Store conditions under which a rule, model or procedure is valid.
24. Knowledge can be general or local
Some knowledge travels broadly; some is tied to one system, location, community or technology stack.
25. Knowledge can be current or historical
An obsolete scientific model may remain valuable for understanding history even if it is no longer accepted as current explanation.
26. Superseded does not mean useless
Old knowledge can show how understanding developed, why decisions were made and which assumptions changed.
27. Certainty is another dimension
Established, probable, provisional, disputed and unknown are knowledge states rather than topic categories.
28. Confidence needs provenance
Why is a claim considered secure? Store method, evidence and authority rather than a bare confidence label.
29. Provenance is part of knowledge architecture
Who asserted, discovered, measured, taught or published the knowledge matters for interpretation and audit.
30. Citation is a relationship, not just formatting
A citation links a claim or artefact to a source and can support provenance, dependency and challenge tracking.
31. Knowledge is naturally networked
Concepts depend on other concepts, claims support explanations, procedures use tools, and models apply to systems.
32. Hierarchy alone is insufficient
Broader-narrower subject trees are useful, but dependency, causation, contradiction and evidence need typed relationships.
33. Facets help organise knowledge
Domain, audience, level, knowledge type, evidence status and time can be separated instead of mixed into one giant taxonomy.
34. Learning sequence is not knowledge hierarchy
What should be learned first is a pedagogical dependency and may differ from conceptual generality.
35. Prerequisite relations deserve explicit edges
A learner may need concept A before B even when A is not the parent category of B.
36. AI can help classify knowledge artefacts
Models can infer topic, difficulty, knowledge type and related concepts, but automated classifications should map to governed concepts and preserve uncertainty.
37. AI-generated statements are claims
They should not be classified as established knowledge merely because they are fluent. Evidence and provenance still matter.
38. A practical knowledge record
- knowledge ID;
- knowledge type;
- canonical concept;
- domain;
- scope;
- claim or content;
- evidence links;
- provenance;
- confidence state;
- prerequisites;
- related concepts;
- valid time;
- audience or level;
- version.
39. Knowledge classification should preserve change
When understanding improves, preserve old models and claims as historical states while updating the current canonical view.
40. The deeper idea
Knowledge is not a warehouse of facts. It is a structured network of concepts, claims, evidence, models, procedures and uncertainties.
To categorise knowledge well is to preserve not only what we think we know, but what kind of knowing it is, why we believe it, where it applies and how it may change.
Final answer
Categorise knowledge by type, domain, scope, evidence status, certainty, provenance, relationships and use. Keep facts, concepts, models, procedures, claims and skills distinct. Preserve uncertainty and historical versions, represent prerequisites and evidence as relationships, and treat AI-generated statements as claims until evidence establishes them more strongly.