Classification is how you carve reality into categories

You already have a classification system. You just haven't examined it.

Right now, without thinking about it, you classify hundreds of things every day. You walk into a meeting and classify people as allies, skeptics, and neutral parties. You open your inbox and classify messages as urgent, routine, and ignorable. You scan a menu and classify dishes as things you'd eat, things you wouldn't, and things you'd try if someone else ordered them.

These aren't conscious decisions. They're classification systems running in the background — sorting reality into buckets that determine what you notice, what you ignore, and what you do next.

Here's the problem: most people never examine the systems they use to carve up the world. They inherit categories from their culture, their profession, their upbringing, and their tools — then treat those categories as features of reality rather than features of their own cognition. A doctor sees a person and classifies by symptom cluster. A marketer sees the same person and classifies by demographic segment. A therapist classifies by attachment style. Same person, three classification systems, three entirely different responses.

Classification is how you carve reality into categories. And the way you carve determines everything downstream — what problems you can see, what solutions you can reach, and what possibilities remain permanently invisible.

Twenty-five centuries of arguing about categories

The question of how to properly categorize things is one of the oldest in human thought, and it's far from settled.

Aristotle launched the Western tradition with what we now call the classical theory of categorization. In his Categories (written around 350 BCE), he proposed that every entity in the world can be classified by a set of necessary and sufficient conditions — properties that something must have to belong to a category, and that are enough to guarantee membership. A triangle has three sides and three angles. That's necessary and sufficient. Everything with those properties is a triangle; nothing without them is. The classical view assumes categories have hard boundaries, that membership is binary (in or out), and that all members of a category have equal status.

This model held for over two thousand years. It's elegant. It's rigorous. And it's mostly wrong — at least for the categories that matter to everyday thinking.

Ludwig Wittgenstein broke the spell in 1953 with a deceptively simple question in Philosophical Investigations: what do all games have in common? Board games, card games, ball games, Olympic games, war games, word games — Wittgenstein examined case after case and concluded that there is no single feature shared by everything we call a "game." Instead, games are connected by what he called family resemblance — a network of overlapping similarities where no one feature runs through them all. Some games involve competition, but not solitaire. Some involve skill, but not roulette. Some are amusing, but not professional chess at the highest level. The category "game" holds together not through a shared essence but through crisscrossing threads of similarity.

Eleanor Rosch made Wittgenstein's philosophical argument empirical. In a series of landmark studies in the 1970s at UC Berkeley, Rosch demonstrated that real human categories have internal structure. When she asked 200 American college students to rate how well various items represent the category "furniture," they consistently rated "chair" as a better example than "lamp," and "lamp" as a better example than "telephone." This means category members are not equal — some are more prototypical than others. A robin is a more prototypical bird than a penguin. Everyone agrees on this, across demographics. Rosch called this prototype theory: categories are organized around central, best-example members, and membership grades outward from there.

Rosch also discovered basic-level categories — the level of abstraction that humans naturally use. You call it a "chair," not "furniture" and not "kitchen chair." You call it a "dog," not "animal" and not "golden retriever." The basic level is where the most useful information clusters, where motor programs live (you know how to sit in a "chair" generically), and where children learn their first category names. This finding, published with Mervis, Gray, Johnson, and Boyes-Braem in 1976, showed that categorization isn't arbitrary — there are levels of abstraction that fit how human bodies and minds interact with the world.

George Lakoff took the argument further in Women, Fire, and Dangerous Things (1987), arguing that categories are not just cognitively structured but embodied — shaped by human perception, motor activity, and culture. The book's title comes from Dyirbal, an Australian Aboriginal language where the noun class that includes women also includes fire, dangerous things, and certain animals. The classification seems absurd by English standards, but it follows coherent internal logic within Dyirbal cultural experience. Lakoff's point: categories don't float free of the beings who create them. They are built from the ground up out of bodily experience, metaphor, and cultural context.

Why classification matters for your thinking

These aren't academic curiosities. The way you classify directly controls what you can think.

Classification determines what questions you can ask. Consider how a hospital classifies patients. If the system classifies by diagnosis, the staff can easily ask: "How many patients have pneumonia?" But if they need to ask "How many patients arrived via ambulance?" and that data wasn't part of the classification schema, the question becomes unanswerable — not because the information doesn't exist, but because the classification system didn't carve reality that way.

This is exactly what happens in database schema design. When engineers model a system, the entity types, relationships, and attributes they choose don't describe an objective reality — they construct a lens. A schema with a customer table that includes a segment column (enterprise, mid-market, SMB) lets you query by segment but collapses all the other ways you might want to slice your customer base. Add a use_case column and suddenly new questions become possible. The schema is a classification system, and it determines what the entire organization can see and act on.

Classification determines what you group together. Jorge Luis Borges illustrated this brilliantly in his 1942 essay "The Analytical Language of John Wilkins," where he attributed to a fictional Chinese encyclopedia a taxonomy that divides animals into: (a) those that belong to the Emperor, (b) embalmed ones, (c) those that are trained, (d) suckling pigs, (e) mermaids, (f) fabulous ones, (g) stray dogs, (h) those included in this classification, (i) those that tremble as if they were mad, (j) innumerable ones, (k) those drawn with a very fine camel's hair brush, (l) others, (m) those that have just broken a flower vase, and (n) those that resemble flies from a distance.

Michel Foucault opened The Order of Things (1966) with this passage, writing that it provoked in him "laughter that shattered all the familiar landmarks of thought." The list is funny because it violates every principle of classification we take for granted. But Foucault's deeper point is that our own classification systems are equally contingent — equally constructed — we just can't see it because we're inside them.

Classification determines what you separate. When you classify your work into "professional" and "personal," certain projects become invisible to certain conversations. A creative side project that could inform your professional work gets filed under "personal" and stays there. When you classify colleagues as "technical" and "non-technical," you miss the product manager who can think about systems architecture or the engineer who understands customer psychology. Every boundary your classification system draws also draws a line of invisibility.

How children build their first classification systems

Jean Mandler's research at UC San Diego on infant cognition reveals something striking about how classification develops. Infants as young as seven months old can distinguish between broad categories — animals versus vehicles, for instance — even before they can name them. But they begin with global categories, not specific ones. A nine-month-old knows the difference between "things that move by themselves" and "things that need to be pushed" before they know the difference between a dog and a cat.

This means classification is not something we learn as an intellectual exercise. It is one of the first cognitive operations the brain performs — a prerequisite for language, for prediction, for survival. You were classifying the world before you could speak. What Mandler's work demonstrates is that our earliest categories are built from spatial and dynamic properties (self-movement versus caused movement, containment versus support) — bodily experiences that become the scaffolding for all later abstraction. Classification doesn't sit on top of cognition. It is the foundation.

How machines classify — and what they reveal about us

When engineers build a machine learning classifier — a decision tree, a random forest, a neural network — they face the same fundamental problem you do: how do you draw boundaries between categories?

A decision tree makes this visible. It takes a dataset and asks: what single feature, split at what threshold, best separates the categories? Maybe it's "age greater than 35" as the first split, then "income greater than $50,000" on the left branch, then "clicked an ad in the past 7 days" on the right. Each split is an explicit classification decision — a line drawn through the data. The resulting tree is a classification system you can read and interrogate. You can ask: why did this item end up in this category? And the tree gives you a traceable answer.

Neural network classifiers achieve higher accuracy but at a cost: the classification logic becomes opaque. The network learns to draw boundaries in high-dimensional space, but no one — not even the engineers who built it — can fully explain why a particular input lands in a particular category. The categories work (the network classifies images of cats and dogs with 99% accuracy) but the classification system is not examinable.

This mirrors a pattern in human cognition. Your conscious classification systems — the ones you can explain and defend — are like decision trees. But most of your classification happens through fast, automatic processes more like a neural network: you classify a situation as "dangerous" or a person as "trustworthy" without being able to articulate the features that drove the classification. The power of building explicit classification systems is that it moves more of your thinking from the opaque, unexaminable mode into the transparent, improvable mode.

The twenty lessons ahead

This is the first of twenty lessons in Phase 12: Classification and Typing. Here is the arc you're entering:

First, you'll confront the fact that categories are constructed, not discovered (L-0222) — there is no objectively correct way to carve up reality, only more and less useful ways. You'll learn why explicit categories beat implicit ones (L-0223) — because you can only improve what you can see.

Then you'll explore the tradeoffs in classification resolution. Binary categories lose information (L-0224) — good/bad, success/failure — while spectrum thinking preserves nuance (L-0225). You'll learn to build taxonomies — hierarchical classification structures (L-0226) — and master the discipline of making categories mutually exclusive and collectively exhaustive (L-0227).

The middle of the phase introduces typing — classification with constraints that prevent errors (L-0228). You'll build practical type systems: status types that track lifecycle (L-0229), priority types that enable triage (L-0230), and role types that clarify relationships (L-0231).

Then you'll face the maintenance costs. Classification debt accumulates (L-0232) when systems aren't maintained. Reclassification is not failure (L-0233) — it's a sign your understanding has grown. You'll learn to calculate the cost of miscategorization (L-0234) and to recognize that classification reveals what you value (L-0235).

The final stretch introduces advanced patterns: prototype-based categories (L-0236) organized around central examples rather than rules, boundary cases that stress-test your systems (L-0237), cross-cutting categories that slice the same data along multiple axes (L-0238), and classification as compression (L-0239) — the fundamental information-theoretic tradeoff beneath all categorization. The phase closes with the principle that good classification systems evolve (L-0240) — they are living infrastructure, not fixed containers.

By the end of Phase 12, you won't just understand categorization as an abstract concept. You'll be able to design, evaluate, and evolve classification systems for your own thinking, your projects, and your tools.

Your Third Brain: classification as computation

Every AI system that processes your data depends on classification. When your email client sorts messages into tabs, it runs a classifier. When your note-taking app suggests tags, it classifies by content similarity. When a search engine ranks results, it classifies pages by relevance. When a large language model responds to your prompt, it has classified your intent, your topic, and your expected output format — all in the first few tokens.

The implication for your cognitive infrastructure is direct: the classification systems in your tools shape what you can think with those tools. A task manager that only supports flat lists forces you to classify by priority or due date. One that supports nested projects, tags, and custom fields gives you a multi-dimensional classification space. Neither is inherently better — but they enable different kinds of thinking.

When you build a personal knowledge system — a Zettelkasten, a digital garden, a project management workflow — you are building a classification system. The folder structure, the tagging scheme, the metadata fields, the links between notes — these are all classification decisions. And every one of them determines what connections you'll find later and which ones will stay hidden.

The most powerful move is to make your classification systems explicit and revisable. Write down your categories. Document why you chose them. Review them on a schedule. Because the moment you treat a classification system as finished, it starts becoming a cage instead of a tool.

Protocol: Audit one classification system

Here is the protocol for putting this lesson into practice:

1. Choose one system. Pick a classification system you interact with daily — your file folders, your task categories, your note tags, your calendar labels, your project statuses.

2. Map its current categories. Write down every category in the system. Include catch-all buckets like "Other" or "Miscellaneous."

3. Identify the original question. What question was this system designed to answer? "What project does this belong to?" "How urgent is this?" "What topic is this about?"

4. Ask the harder question. Is that still the most important question? What question do you actually find yourself struggling to answer? If you keep searching for things and not finding them, your classification system is optimized for the wrong query.

5. Find three items that don't fit. Look for items that sit awkwardly between categories, that could go in multiple buckets, or that you've put in "Other." These boundary cases reveal where your classification system is breaking down.

6. Sketch one alternative. Design a different classification system for the same items — one optimized for the question you actually need answered. You don't have to implement it. Just the act of designing an alternative breaks the illusion that your current system is "the way things are."

Classification is not something you do once. It is infrastructure you build, maintain, and evolve. The quality of your categories determines the quality of your thinking. And the first step to improving your categories is recognizing that you chose them — even the ones you inherited without noticing.