The Question Space Is Larger Than the Knowledge Space

Digital publishing contains a fundamental asymmetry:

Humans can ask vastly more questions about a domain than there are underlying facts required to answer them.

The distinction seems obvious once stated. Its consequences are not.

Publishing has traditionally organized itself around information demand. Publishers discover what people want to know and create information products intended to satisfy that demand.

2.1 Questions Multiply Faster Than Facts

Consider a product represented by ten useful attributes.

Those ten attributes do not support only ten questions.

A customer can ask directly about any one of them. What does the product cost? How much does it weigh? What is its capacity?

The attributes can also be compared.

Which product is lighter?

They can be filtered.

Which products cost less than $500?

Conditions can be combined.

Which products cost less than $500 and weigh less than two kilograms?

Results can be ranked.

Which qualifying product is cheapest?

Values can participate in calculations.

How much lighter is Product A than Product B?

Relationships introduce additional possibilities. So do geography, availability, configuration, and time.

Then every underlying information need can be expressed in many different ways.

The number of facts has not grown nearly as quickly as the number of questions those facts can support.

Add another useful attribute and it does not merely add another possible question. It can participate in combinations with the attributes already represented.

The effect compounds as the domain becomes richer.

Question Space grows combinatorially. Knowledge Space does not.

That matters because it suggests that information demand and publishing output do not necessarily need to grow together.

A new question may require new knowledge.

But it may also be nothing more than a new request across knowledge the publisher already possesses.

The difference becomes easier to see with a concrete example.

2.2 The Camera Catalog Example

Consider a photography retailer selling 2,000 cameras.

For every camera, the retailer maintains a consistent set of information: manufacturer, model, camera type, sensor format, resolution, weight, dimensions, lens mount, stabilization, video capabilities, and price.

The retailer has not attempted to anticipate every question a customer might ask about those products. It simply maintains authoritative and consistently represented product information.

A consumer asks:

Which full-frame mirrorless cameras under $2,500 weigh less than 700 grams and have in-body stabilization?

The question introduces several conditions:

camera type: mirrorless
sensor format: full-frame
price: less than $2,500
weight: less than 700 grams
stabilization: in-body

The retailer did not need to predict that combination.

No keyword researcher needed to discover the query. No editor needed to commission an article titled Best Lightweight Full-Frame Mirrorless Cameras Under $2,500 With IBIS. No writer needed to create the answer. No page needed to exist for that precise information need.

The necessary knowledge was already present in the catalog.

The question supplied the projection.

Change the ranking criterion, manufacturer, lens mount, weight differential, or another available dimension and the publisher still does not necessarily need new source knowledge.

The projection changes.

The underlying knowledge does not.

This depends, however, on the information being represented consistently.

If the 2,000 cameras are described independently in prose, a consuming system may first need to determine which specifications are comparable. One manufacturer may use “image stabilization,” another “5-axis sensor-shift stabilization,” and another “IBIS.” One weight measurement may include the battery while another excludes it. Different configurations may be identified inconsistently.

Before the facts can be compared, somebody has to establish what those facts mean relative to one another.

If the publisher has already done that work, it does not need to be reconstructed independently for every future question.

The publisher establishes comparability before the question arrives.

That changes what the publisher needs to manufacture.

The retailer does not need 50,000 prewritten answers to 50,000 anticipated questions. It needs reliable knowledge about the 2,000 cameras.

The knowledge existed.

The answer did not.

The answer could be constructed when the question arrived.

Real domains are more complicated than a camera catalog. Facts can depend on time, geography, identity, terminology, relationships, eligibility, or other conditions. We will encounter those problems later.

They do not change the underlying asymmetry.

The possible questions multiply much faster than the authoritative knowledge from which they can be answered.

2.3 The Traditional Response: Publish the Question Space

Digital publishing has historically responded to new information demand by creating new content.

That strategy was rational when the page was the information product and the destination where the user’s task was expected to be completed.

But it has an architectural consequence.

The same underlying knowledge gets published repeatedly.

One fact may appear on a primary resource, several explanatory articles, comparison pages, FAQs, location pages, buying guides, and programmatic resources. Conceptually, the organization may possess one authoritative fact while operationally maintaining many expressions of it.

The problem becomes more visible as a publishing operation grows.

A subject such as Medicare dental coverage, for example, can produce article after article addressing slightly different formulations of substantially overlapping information needs. Each new artifact may have a legitimate editorial purpose. Collectively, however, they can reproduce the same underlying facts across a growing corpus.

Every additional expression creates another place where information can become stale, inconsistent, contradictory, detached from its provenance, or simply expensive to maintain.

This is not necessarily poor editorial practice.

It is a predictable consequence of attempting to publish the Question Space.

The same pattern can persist even when publishing adapts to generative search. A 1,500-word article may become a paragraph, FAQ response, table, snippet, or another small retrievable unit.

The artifact has become smaller.

But if every anticipated question still produces another independently maintained answer, the underlying model remains:

Question → Answer Artifact

The difficulty is that the Question Space has no practical boundary.

Users can continuously combine known dimensions in ways the publisher did not anticipate. They can change the comparison, add a condition, change the geography, alter the time period, traverse another relationship, or ask the same underlying question in language nobody inside the publishing organization ever researched.

No publisher can manufacture every legitimate permutation of a sufficiently rich domain.

Increasingly, it may not need to.

2.4 Knowledge-Space Publishing

There is another place to invest the publishing effort.

Instead of attempting to manufacture the Question Space, the publisher can maintain the smaller Knowledge Space beneath it.

When a new question can be answered from that knowledge, the question defines the projection.

When it cannot, the question has revealed something the publisher does not yet know, has not represented, or cannot safely determine.

The two approaches have very different shapes:

Question-Space Publishing

Question → Answer Artifact
Question → Answer Artifact
Question → Answer Artifact
Question → Answer Artifact

Knowledge-Space Publishing

             Question
                ↓
Question → KNOWLEDGE ← Question
                ↑
             Question

This changes the economics of publishing.

Under a Question-Space model, supporting additional information needs generally means producing additional artifacts. More questions create more things to write, publish, update, validate, and maintain.

Under a Knowledge-Space model, improving the maintained knowledge can increase the number of information needs that the existing knowledge can support.

A new attribute can participate in combinations with attributes already known.

A new relationship can make new questions possible across existing entities.

A correction to an authoritative fact can improve future answers that depend upon it without requiring the publisher to locate every independently authored expression of the old value.

Knowledge can therefore acquire compounding utility.

The objective is not simply to produce answers more cheaply.

It is to increase the range of legitimate information needs that can be supported by a maintained body of authoritative knowledge.

That leads to a governing principle for the architecture developed in this book.

2.5 Engineer Once. Answer Many.

Engineer once. Answer many.

This does not mean that every information need can or should be reduced to facts and computation.

Human communication does things that factual projection cannot.

A photographer may want to understand why one camera system is better suited to wildlife photography than another. Specifications alone may not provide the answer. Experience, tradeoffs, and judgment matter.

Write that.

A Medicare beneficiary may need to understand the practical consequences of choosing between Original Medicare and Medicare Advantage. Facts alone may not provide sufficient explanation.

Explain that.

Some information needs require analysis. Others require instruction, interpretation, narrative, persuasion, or judgment.

Knowledge-Space publishing does not eliminate those forms of publishing.

It distinguishes them from another class of information need: questions that ask for a new projection of knowledge the publisher already possesses.

That gives the publisher a useful question to ask before creating another answer artifact:

Does this information need require new knowledge—or merely a new projection of knowledge already possessed?

If it requires explanation, create the explanation.

If it requires judgment, provide the judgment.

If the question exposes something the publisher does not know, improve the underlying knowledge where appropriate.

But if the necessary knowledge already exists, another independently maintained answer may not be necessary.

Let synthesis do what synthesis is good at.

The Question Space will continue to expand. People will combine known facts in new ways, introduce new conditions, request new comparisons, and ask questions publishers never predicted in language publishers never researched.

The publisher does not need to anticipate all of them.

Engineer the Knowledge Space.

Let the question arrive later.

That raises the next problem.

If the durable asset is knowledge rather than an ever-growing collection of answers, what makes something knowledge rather than merely data?