the second hundred · metaphor 145
Look at almost any friend group, lunchroom, or neighbourhood and it has quietly sorted — like sitting with like. It is tempting to look for who arranged it. Often nobody did.
Nobody in the crowd wants a segregated world. Each person wants something small and forgivable: not to be the only one of their kind on the block, a couple of familiar faces at the table, a sense that they aren't completely surrounded by strangers. These are mild wishes, held privately, with no coordination between them. You would swear that turning them into sharp, sorted clusters would take an organiser — a policy, a gatekeeper, a hand on the scale.
It doesn't. A very weak preference — a wish that merely a third of your neighbours resemble you — is enough, once everyone acts on it and people are free to move, to tip a thoroughly mixed world into a starkly divided one. The unsettling part is the gap between cause and effect: gentle, blameless individual taste at one end; hard collective segregation at the other; and no segregator anywhere in between.
The tipping model
Thomas Schelling asked, in 1971, what would happen if each person tolerated being a minority on their block only down to some threshold — and moved when they slipped below it. His startling answer, worked out with coins on a checkerboard, was that thresholds far below any wish for actual separation still drove near-total separation. A population content to be a one-third minority — happy so long as it isn't badly outnumbered — sorts itself almost completely.
The engine is iteration. One discontented person moves, which nudges the make-up of the block they leave and the block they join, which tips a formerly-content neighbour of theirs past their threshold, who then moves too. Small individual adjustments compound. There is no step at which anyone chooses segregation; there is only a long chain of locally reasonable moves whose global sum is a sorted map. Sociologists call the underlying pull homophily — the tendency of like to associate with like — and Schelling's grid shows what homophily does when you let it run.
The segregation index makes it legible: for every resident, the share of their occupied neighbours who match their kind, averaged over everyone. Sprinkle two groups at random and it sits near 0.5 — a coin-flip neighbour. As the grid sorts, it climbs toward 1.0, where almost everyone is surrounded by their own. Nothing sets that number by hand; it is read off the map each round.
What to try
Leave the slider at 0% and press run: nobody is ever discontented, nobody moves, the index stays at its mixed 0.5. Now raise it a little. At a wish for just one-third like-neighbours the map still churns for a while and settles far more sorted than it began — the index climbs from ~0.50 to somewhere near 0.75, and every resident ends up content. Nobody got what a bigot would want; everybody's modest wish was met; the neighbourhood segregated anyway.
Push the slider higher and the effect runs away — at half or more, the index heads for the high 0.9s and the map splits into two solid territories. The lesson is in the disproportion, not the extremes: watch how a small move of the slider produces a large move of the index. The sorting is not proportional to the prejudice. A faint preference and a strong one produce a similar-looking wall; the difference between a mixed world and a divided one can be a threshold nobody would even confess to holding.
The mapping
Read the grid as any social space where people can drift toward comfort — a cafeteria, a group chat, a city, a feed, a profession. The two colours are whatever line people sort along: background, politics, taste, class, first language. The threshold is the private, usually unspoken level of same-ness each person needs to feel at home. Nobody sets out to build an enclave. Each just gravitates, a little, toward tables and timelines where they aren't the odd one out — and the enclave assembles itself behind their backs.
This is why the sorted world is so easy to misread. Standing inside a homogeneous cluster, it is natural to infer strong preferences, or a designing hand — someone must have wanted this. The model warns against both inferences. The outcome licenses no conclusion about the strength of anyone's taste, and none about intent. A neighbourhood, a party, a school clique can be intensely segregated and composed entirely of people who would each have happily lived mixed. Structure is not a confession.
Read as life lessons
Individually reasonable moves can add to a collectively awful outcome. There is no rung on the ladder where the micro-motive looks like the macro-result. Judge systems by what they produce, not only by what each actor intended.
A pattern needs no author. "Who did this?" can have the honest answer no one — which is precisely why it's so hard to undo. You can't fire the person in charge of a feedback loop.
A divided map is not proof of strong preference or hidden malice. Weak taste plus free movement plus time is enough. Reading intensity of feeling off the result is exactly the mistake the model exposes.
In the wild
Residential segregation persists in places with no legal barrier, driven partly by exactly this mechanism — though real housing also carries the heavier weights of income, discrimination and policy the toy grid leaves out.
McPherson's “birds of a feather” review found homophily structuring who we befriend, marry and talk to. Online, low-friction sorting builds echo chambers fast: it is cheaper to unfollow than to move house.
Professions, subfields and hobbyist scenes drift homogeneous by the same slow pull — each newcomer nudged toward the room where a few people already look or think like them.
The mapping, exactly
| Mathematics | Life |
|---|---|
| the two colours | Any line people sort along — background, politics, class, taste, tongue. The model doesn't care which. |
| the threshold τ | The private, unspoken level of same-ness a person needs to feel they belong somewhere. |
| a discontented agent | Someone who feels like the odd one out — and, being free to, drifts toward a more familiar room. |
| iterating the moves | Time: each small relocation reshuffles who's around whom, tipping the next person, and the next. |
| segregation index → 1 | The finished enclave — nearly everyone surrounded by their own, though no one aimed for it. |
| no central rule | The absent segregator: a hard pattern with no author, which is why blame and repair both slip. |
The honest model
The grid is a real neighbourhood of two kinds of resident with a scatter of empty cells. A resident is content when the share of their occupied Moore neighbours (the eight surrounding cells) who match their own kind is at least the threshold τ you set with the slider; otherwise they are looking to move. Each round, everyone discontented relocates, in random order, to a randomly chosen empty cell — the plainest version of Schelling's rule, no smart destination-seeking.
The segregation index is measured, not assigned: every round the code visits each resident, computes their matching-neighbour fraction, and averages. That is the number in the meter and the climbing spark. Content and looking-to-move counts are likewise tallied from the live grid. Set τ to zero and no one is ever discontent, so nothing moves and the index holds at its random baseline — the model's own proof that the sorting is driven by the preference, not by the shuffling.
Where the metaphor tears
The model's power is to show sorting can arise from blameless taste alone. It is abused the moment that becomes an alibi. Real-world segregation is also produced by money, redlining, discrimination and law — active segregators with names. Homophily is one mechanism among several, not an exoneration of the others.
Each resident has one immutable colour, and preference points only from company toward similarity. Real life runs the other arrow too: we become like those we're near. When identity and influence entangle — you take on your neighbours' tastes as much as you seek your own — the clean story of pre-sorted types choosing sides breaks down.
The toy lets a discontented agent teleport to any empty cell at no cost. Actual relocation — of home, job, or social world — is expensive, slow, and constrained, which damps the cascade and can freeze a mixed-but-unhappy arrangement in place. The mechanism is real; its speed and reach in the model are not.