The Map Is Not the Territory—and the Dataset Is Not the Place

“The map is not the territory” is usually used as a warning about cartography. The same warning should be applied one step earlier.

The dataset is not the place either.

A geographic dataset is a structured account of a place—selected features, chosen boundaries, measured attributes, dates, definitions and omissions—and can be excellent data while still representing only one useful version of the world.

That distinction becomes easy to forget because modern mapping interfaces make data feel immediate. Pan across a city, click a polygon, inspect an exact coordinate—the experience can feel like interacting with the place itself.

What you are interacting with is a model.

The world is continuous; datasets are selective

A road dataset may contain classified roads and omit informal tracks. A hospital registry may include licensed facilities and exclude temporary clinics. A forest layer may define forest according to canopy threshold, minimum area and observation date. A city polygon may represent a legal municipality rather than the built-up urban area people recognise as the city.

These choices do not make the datasets defective. They make them purposeful.

Every dataset has a theory of what counts.

Trouble begins when that theory disappears during reuse: a legal boundary becomes “the city”, a point returned by a gazetteer becomes “the village”, and a land-cover class becomes “what the land is”. In each case, the representation is promoted into the thing represented.

Detail can hide abstraction

High geometric detail creates a particular kind of confidence.

A polygon with thousands of vertices looks precise, a coordinate with six decimal places looks exact, and a high-resolution raster looks close to reality.

But technical detail does not answer whether the feature represents the intended phenomenon.

A district boundary can be geometrically accurate and analytically irrelevant to a functional labour market, while a precise centroid can be the wrong model for a dispersed settlement. Even a one-metre image can still be several years old or classified with the wrong categories.

Precision is a property of representation. Fitness is a relationship between representation and question.

Derived geography increases the distance

Analysis often creates new geographic objects that never existed as observations.

A buffer is a rule applied to a geometry. An isochrone is a model of reachable space. A hotspot is an analytical construct. A harmonised boundary is an attempt to make changing geographies comparable. An extracted feature may combine documentary evidence with a geocoding decision.

These derived objects can be extremely useful, but they require a different kind of trust because their meaning depends not only on where they are but on how they were produced.

A 30-minute service area is not “the area people can reach”. It is the area reachable under a particular network, routing profile, origin, direction, time assumption and set of constraints. The geometry makes the result usable; provenance makes it interpretable.

Naming is another form of modelling

Even labels are representations.

Place names can vary by language, institution and period. Administrative codes change. A locality may have several accepted names. A feature labelled “forest” may satisfy one institution's definition and fail another's.

The dataset needs identifiers and categories because analysis requires structure. The mistake is treating those structures as natural facts rather than conventions created for a purpose.

Once that is understood, disagreements become easier to reason about. Two datasets can describe the same place differently without one being obviously corrupt. They may be answering different questions.

Reuse requires context, not just files

The farther data travels from its producer, the more important its context becomes.

A future user needs to know what the features represent, which date applies, which definitions were used, where the data came from and what transformations have already occurred. Otherwise the next map begins by inheriting assumptions it cannot see.

This is why reusable geographic data is not merely data in an open format. Reuse depends on preserving enough of the model to decide whether it still fits a new task.

That does not require philosophical doubt around every point and polygon. It requires a practical habit: ask what the feature stands for before asking what can be done with it.

Good geographic work respects the gap

Maps and datasets are valuable precisely because we cannot carry the territory into the analysis room.

Abstraction is not the problem; unacknowledged abstraction is.

The mature question is therefore not “Is this dataset real?” It is:

What part of the place does this dataset represent, by what method, for what time, and for which decisions?

That question keeps the representation useful without asking it to be the world itself.