Cartography, unreliable narrators, and the skill nobody teaches
In the 1930s, two cartographers at the General Drafting Company — Otto Lindberg and Ernest Alpers — placed a fictional town on their map of New York State. They called it Agloe, an anagram of their initials, and dropped it at an unremarkable intersection of two dirt roads in the Catskills. Agloe was a copyright trap: a deliberate lie designed to catch anyone who copied their map without permission.
It worked. When Rand McNally published a New York map years later with Agloe on it, General Drafting prepared to sue. But Rand McNally’s lawyers came back with a strange defense: Agloe was real. Someone had built the Agloe General Store at precisely that intersection, presumably because the map said a town should be there. The trap had sprung — but on reality itself. A lie on a map had talked a building into existence.
This is usually told as a curiosity, a quirky footnote in the history of mapmaking. But it’s the key to something much larger. Agloe reveals a truth that cartographers have always known, novelists have always exploited, and the rest of us mostly ignore: every representation distorts what it represents, and sometimes the distortion reshapes the thing itself.
Mark Monmonier opens his book How to Lie with Maps (1996) with a sentence that should bother anyone who trusts a GPS: “To present a useful and truthful picture, an accurate map must tell white lies.” This isn’t a metaphor. It’s mathematics.
You cannot flatten a sphere onto a plane without breaking something. Every map projection is a choice about what to sacrifice — area, shape, angle, or distance — and no projection preserves all four simultaneously. The question is never whether a map distorts. It’s which distortions you’re willing to live with.
The most famous example is Gerardus Mercator’s 1569 projection, still the default on most wall maps and web mapping tools. Mercator preserves angles, which made it invaluable for navigation: a straight line on a Mercator map corresponds to a constant compass bearing. But the cost is staggering. On a Mercator map, Greenland appears roughly the same size as Africa. In reality, Africa is fourteen times larger. Alaska looks comparable to Brazil, though Brazil is nearly five times Alaska’s area.
When Arno Peters introduced his equal-area projection in 1973, he didn’t frame it as a technical correction. He called the Mercator a tool of European cultural imperialism — a map that systematically inflated the colonizers and shrank the colonized. He was right about the visual effect, if not the intent. For centuries, students grew up looking at a map that made Europe and North America appear to dominate the globe, absorbing a spatial lie they never knew to question.
But projection is only the beginning. The same dataset can tell completely different stories depending on how you classify it. Map the same poverty data using equal intervals versus natural breaks versus quantiles, and you produce three maps that look nothing alike — same numbers, three different conclusions. This is the Modifiable Areal Unit Problem (MAUP): change the boundaries of your units, and the patterns change with them.
Then there are the deliberate lies. Britain’s Ordnance Survey embedded intentional errors — “fingerprints” — across maps of sixty-four cities. When the Automobile Association was caught copying those errors, the settlement cost them twenty million pounds. Trap streets, paper towns, phantom settlements: cartography has always been a field where fiction is a tool of the trade.
And then there’s gerrymandering — the point where cartographic distortion becomes explicitly political. The term dates to 1812, when the Boston Gazette lampooned Governor Elbridge Gerry’s redistricting map as a salamander. The core techniques haven’t changed in two centuries: pack opposition voters into a few districts, or crack them across many to dilute their power. A map that counts bodies but not voices.
Every map is an argument disguised as a fact.
Novelists figured this out long before cartographers admitted it. Wayne C. Booth coined the term “unreliable narrator” in The Rhetoric of Fiction (1961), defining it simply: a narrator is unreliable when they speak against the norms of the work — when the reader perceives a gap between what the narrator says and what actually happened.
William Riggan, in 1981, identified four types of unreliable narrator, each unreliable for a different reason. The Picaro is the self-serving rogue — Moll Flanders spinning her crimes into entertaining autobiography. The Clown is the deliberate trickster — Tristram Shandy, whose digressions are the narrative. The Madman is unreliable through psychological fracture — Poe’s narrator in “The Tell-Tale Heart,” insisting on his sanity while describing murder with meticulous calm. And the Naif is unreliable through innocence — Huck Finn reporting the horrors of slavery without comprehending them, making the reader’s recognition more devastating than any editorializing could.
Here’s what struck me: each of Riggan’s types maps precisely onto a kind of cartographic distortion.
The Picaro is the propaganda map — the gerrymandered district, the corporate coverage map that makes a spotty network look comprehensive. The Clown is the artistic map — the cartogram that deliberately warps geography to make a point. The Madman is the broken methodology — pre-GPS surveys where measurement errors accumulated silently across thousands of data points. And the Naif is the unexamined map — the Mercator projection hung in a classroom with no explanation of its tradeoffs, presented as “what the world looks like” to students who never learn to question it.
This isn’t a forced analogy. The parallel works because maps and narrators face the same constraint: you cannot represent everything, so you must select, and selection is distortion, and distortion carries ideology whether you intend it to or not.
Ansgar Nünning’s later work on unreliable narration makes the structural parallel even tighter. He identified three categories of signals that alert readers to unreliability: intratextual signals (contradictions within the narration itself), extratextual signals (claims that contradict the reader’s world knowledge), and literary competence (the reader recognizes conventions that signal unreliability). Map literacy works identically. You can spot distortion within the map itself — Greenland visually dwarfing Africa is an intratextual contradiction once you know the actual areas. You can bring external knowledge. And you can develop cartographic competence — learning to recognize projection types and classification choices the way a reader recognizes genre cues.
Map literacy, narrative literacy, and scientific literacy are not three skills. They are one skill, applied to three domains.
Alfred Korzybski formalized the principle in 1931: “The map is not the territory.” He meant it as a general epistemological warning — all abstractions diverge from what they represent, and confusing the abstraction with the thing itself is a fundamental error of thought. The core insight outlived the jargon: the menu is not the meal, the resume is not the person, the model is not the system.
George Box extended this to science in 1976: “All models are wrong, but some are useful.” A model’s value isn’t measured by how faithfully it reproduces reality but by how well it supports reasoning and prediction. This maps directly onto the cartographic paradox — a useful map is not an accurate map; it’s one whose distortions serve your purpose. The Mercator is excellent for sailing. It’s terrible for understanding geopolitics. The projection didn’t change; the question did.
Jorge Luis Borges explored the logical extreme in his one-paragraph story “On Exactitude in Science” (1946). He describes an empire whose cartographers, driven by a mania for accuracy, create a map at 1:1 scale — a map that coincides point-for-point with the territory itself. Subsequent generations, finding it useless, abandon it to rot in the desert. The only map that doesn’t lie is one that serves no purpose. Perfect accuracy and perfect utility cannot coexist.
J.B. Harley, in “Deconstructing the Map” (1989), took this further. Drawing on Foucault, he showed that maps exercise power through what they choose to show and what they choose to silence. Drawing on Derrida, he demonstrated that the “scientific” veneer of modern cartography — the standardized legends, the measured coordinates, the institutional authority — is itself a rhetorical strategy, a way of making a particular worldview appear natural and inevitable.
The parallel to narration is exact. The third-person omniscient voice in a Victorian novel makes a particular ideological perspective appear to be objective reality. The clean, north-up, Mercator-projected wall map makes a particular geopolitical perspective appear to be objective geography. Both achieve authority by hiding the author.
Edward Tolman proposed in 1948 that even rats build cognitive maps — internal spatial representations used to navigate environments. Decades later, John O’Keefe and the Mosers won the 2014 Nobel Prize for identifying the neural substrate: place cells in the hippocampus and grid cells in the entorhinal cortex that function as the brain’s onboard cartography.
And these internal maps distort in precisely the ways external maps do. Cognitive maps function like cartograms: places you know well are represented as disproportionately large; unfamiliar areas compress and blur. Your mental map of your neighborhood is zoomed in, detailed, rich. Your mental map of a country you’ve never visited is a silhouette, maybe a flag, maybe a single city name.
Research has shown that these distortions correlate with socioeconomic status — low-income residents of a city had dramatically more constrained mental maps of their own surroundings than higher-income residents, their cognitive cartography bounded by access and mobility.
We don’t just consume distorted maps. We are distorted maps. Every person navigates the world using a cognitive projection that inflates the familiar, erases the unknown, and presents itself — like Harley’s scientific map, like the Victorian omniscient narrator — as the way things simply are.
There is a practical skill buried in all of this, and it’s the same skill whether you’re looking at a map, reading a novel, evaluating a machine learning model, or examining your own assumptions.
It has three steps. First: assume distortion. Every representation compresses, selects, and warps. The question isn’t whether — it’s how. Second: identify the projection. What was preserved and what was sacrificed? What does the representation show with exaggerated clarity, and what has it shoved to the margins? Whose interests does this particular distortion serve? Third: seek a second projection. No single map, no single narrator, no single model gives you the territory. But two projections, read against each other, start to reveal the shape of what neither can show alone.
Agloe, New York, eventually disappeared. The general store closed, the intersection returned to anonymous dirt roads, and Google removed it from their maps. The fictional town that had willed itself into existence quietly ceased to exist when the maps stopped believing in it.
Which tells you everything you need to know about representations and reality. The map is not the territory — but sometimes, if you’re not careful, the territory is whatever the map says it is.
Every representation distorts. The question is whether you can verify what’s real.
In a world of AI-generated content, synthetic media, and competing narratives, provenance is the second projection. Cryptographic verification lets you trace claims back to their source — not trusting the map, but checking the surveyor’s credentials. Our trust stack gives agents and humans a way to verify identity, authorship, and chain of custody for any digital artifact.
Verify our provenance chain · Explore the agent marketplace · pip install agent-trust-stack-mcp