Chapter 3 left us with a problem.
A fact does not become useful knowledge merely because it is machine-readable. Enough of its meaning must remain with it: what the fact describes, the context in which it applies, its relationships to other things, where it came from, and any conditions governing how it can safely be used.
But meaning also requires organization.
Those elements cannot simply exist somewhere within the publisher’s information environment and be expected to remain associated correctly. Some facts belong together. Others do not. Some context governs many assertions. Some applies to only one. Relationships may connect otherwise separate subjects. Provenance may establish the authority of one fact or an entire collection.
If the publisher is going to engineer the Knowledge Space, it needs some way to determine what knowledge belongs together.
It needs a boundary.
4.1 Knowledge Needs a Boundary
The Web already gives publishers several useful boundaries.
The page is one of them.
A page can be created, named, addressed, linked, indexed, updated, and retrieved. It provides a practical container for human communication.
But the page is primarily a presentation boundary.
One page can contain information about several things. A single subject can require information drawn from several sources. Knowledge represented on one page may depend on relationships or definitions established elsewhere. A page may also contain navigation, advertising, interface elements, explanation, and other material that serves the document without defining the knowledge itself.
The URL tells us where the document lives.
It does not necessarily tell us what knowledge belongs together.
The paragraph provides another useful boundary, but for a different purpose. Paragraphs organize prose. A single paragraph may contain several assertions. One assertion may depend on context established under a heading several paragraphs earlier.
The paragraph is a prose boundary.
The query has a boundary too. It represents an information need. But Chapter 2 showed why organizing knowledge around queries creates a different problem: many questions can project across the same underlying knowledge.
The query describes demand.
It does not define the knowledge beneath that demand.
Even the database row is not necessarily the right boundary. A row reflects the architecture of the source system. One meaningful subject may require records from several tables, datasets, or systems. One source record may also contribute knowledge to several different uses.
The database row is a source-system boundary.
All of these boundaries are useful.
None necessarily answers the question we now need to answer:
What belongs together because it describes the same meaningful thing or state?
That is a semantic question.
Meaning needs its own boundary.
4.2 The Knowledge Object
For the purposes of this book:
A knowledge object is a coherent and bounded representation of something the publisher knows.
The definition is intentionally simple.
Coherent means that the knowledge belongs together because of what it represents. The boundary is not created merely because the information happens to appear on the same page, originate in the same dataset, or answer the same question.
Bounded means that the object represents something sufficiently identifiable that we can distinguish what belongs inside the object from what belongs outside it.
And something the publisher knows establishes the most important characteristic of the object.
It represents knowledge rather than an anticipated answer.
Consider a camera.
A publisher may know the camera’s manufacturer, model, product identifier, sensor format, resolution, weight, dimensions, lens mount, stabilization capabilities, video features, price, and relationships to compatible equipment.
Those facts can form a coherent object because they describe the same identifiable product.
The camera provides the semantic center.
Now consider something less obvious:
2026 Medicare coverage options in Mohave County, Arizona.
There is no single product at the center of that knowledge.
The publisher may know which Medicare Advantage plans are available, which prescription drug plans are offered, which carriers participate in the county, what plan types are represented, what premiums apply, and other facts describing the local Medicare environment.
Those facts may originate in different datasets. They may describe different programs and different entities.
Yet they can still form a coherent knowledge object because collectively they describe the same meaningful state:
Medicare coverage options available in Mohave County during the 2026 plan year.
The boundary comes from meaning.
This distinction is important because a knowledge object is not simply another name for an entity record.
Some knowledge objects may center on identifiable entities such as a product, plan, organization, facility, or provider.
Others may center on a contextual state, a concept, a relationship, or another coherent piece of reality the publisher needs to represent.
The architecture does not require us to create a taxonomy for every possible kind of object.
The governing question is simpler:
What does this knowledge represent?
If the answer is clear, we have the beginning of a semantic boundary.
If the answer requires several unrelated subjects joined only because they happen to appear together, the boundary probably needs more work.
4.3 The Semantic Contract
Once a knowledge object has a boundary, that boundary establishes an implicit contract about what the object means.
The object effectively tells a consumer:
This is what I represent.
Its identity establishes the subject or state around which the knowledge is organized.
This is what I assert.
Its facts establish what the publisher knows within that boundary.
This is the context governing those assertions.
Geography, time, population, configuration, eligibility, jurisdiction, or other conditions may determine where the assertions remain applicable.
These are the relationships that matter.
The object may depend on connections to other entities, concepts, or knowledge outside its own boundary.
This is where the knowledge came from.
Provenance establishes the sources and authority behind what the publisher is asserting.
These are the limits of what can safely be concluded from it.
Some knowledge can be applied directly. Other knowledge may require additional context before one of several valid possibilities can be selected.
These are the dimensions of meaning identified in Chapter 3. The knowledge object gives them a semantic home.
This does not mean every object must state every piece of meaning independently.
Some context may govern the object as a whole. Some may belong to a particular assertion. A relationship may connect the object to something represented elsewhere. Provenance may apply broadly to a collection of facts or specifically to one derived value.
Nor does a useful semantic boundary require the object to become as small as possible.
If every individual fact becomes its own object, the context that makes those facts meaningful may have to be reconstructed repeatedly. At the opposite extreme, placing everything a publisher knows about an entire domain into one object destroys the distinctions that made the knowledge coherent in the first place.
The useful boundary lies somewhere between meaningless atomization and indiscriminate aggregation.
A practical test is:
What piece of reality does this object describe?
The answer should be specific enough to establish what belongs within the boundary, but broad enough to preserve the context required for the knowledge inside it to remain coherent.
That is the semantic contract.
It does not tell the consuming system what answer to produce.
It tells the system what the publisher is representing.
4.4 The Knowledge Object Is Upstream of the Answer
The most consequential property of the knowledge object is not what it contains.
It is where it sits in the publishing process.
A knowledge object is not the query.
It is not the page.
It is not the database row.
And it is not the answer.
Consider the camera again.
The camera knowledge object does not exist solely to answer:
How much does this camera weigh?
Weight is one assertion within what the publisher knows about the camera.
The same object may later participate in a question about price, stabilization, sensor format, lens compatibility, video capability, or a comparison with another camera.
Those questions do not define the object.
The camera does.
The knowledge can therefore exist before any of those questions are asked.
This reverses the sequence established by traditional Question-Space publishing.
In an answer-oriented model:
Question anticipated → Answer authored → Artifact published
In a knowledge-oriented model:
Knowledge represented → Question arrives → Knowledge projected → Answer constructed
The knowledge object occupies the position upstream of the answer.
That connects directly to the asymmetry developed in Chapter 2.
If many questions can project across a comparatively bounded Knowledge Space, then the publisher needs a durable unit within that Knowledge Space that exists independently of those questions.
The knowledge object provides that unit.
The page can present knowledge.
A machine can retrieve knowledge.
A question can request a particular projection of knowledge.
An answer can be constructed from knowledge.
But:
The knowledge object represents what is known.
That distinction changes what the publisher considers the durable asset.
An answer artifact has utility for the information need it was created to satisfy. A knowledge object can potentially participate in many information needs because its boundary is defined by what the publisher knows rather than by the question the publisher anticipated.
It can be updated when reality changes.
It can participate in comparisons.
It can connect to other knowledge through relationships.
It can be represented differently when different consumers require different forms.
And it can support questions that did not exist when the object was created.
This does not eliminate documents or answers. It puts them in a different relationship to the knowledge beneath them.
The document is one way knowledge can be presented.
The answer is one way knowledge can be used.
Neither has to define the knowledge itself.
The Web has always given publishers a unit for documents.
What we had been missing was a unit for knowledge.
Once that unit exists, the publisher no longer has to organize everything it knows around questions people have already asked.
The knowledge can exist first.
The question can arrive later.