spread & inequality · metaphor 50 of 100
The rich get richer — not as a moral failing or a conspiracy, but as a mechanism: wherever new arrivals attach to whatever is already attached-to, early small advantages compound into permanent structure. The mechanism doesn't know who deserves. It only knows who already has.
Two coffee shops open the same week, equally good. One catches a lucky first month — a warm review, a knot of regulars visible in the window. New customers choose partly by crowd, because a crowd is information; so the crowded one crowds further and the empty one empties. A year later one is an institution and one is a for-lease sign, and the whole neighborhood explains the difference by quality.
Citations, followers, fortunes, first names, band fame: the same shape keeps assembling wherever arrival meets attachment — a few giants and a vast thin tail, grown from nothing but join what's joined. Below, the process runs live. Watch the giants condense; then rewind history and watch different giants condense from identical rules.
Honest simulation: 360 nodes arrive one per tick, each attaching m = 2 edges with probability ∝ degree × qualityβ (β = 0 is exact Barabási–Albert). Every attachment is sampled live; every number on this page — slopes, shares, champions — is computed from the run you are watching. Nothing is pre-scripted. Hover or tap a node to interrogate it.
The mechanism
No villain is required — only three ordinary prudences. Visibility: you cannot choose what you cannot see, and what you see — the top of the search page, the front of the reading list, the queue out the door — is what has already been chosen. Vetting: the crowded café has been inspected a thousand times for free; picking the popular thing outsources your due diligence to everyone who came before you. Infrastructure: the big node is simply easier to attach to — the classic paper already sits in every bibliography, the big airport already flies where you're going. Each motive is locally rational, even wise. Summed over arrivals, they assemble a machine.
The machine has one moving part: the probability that a newcomer attaches to you is proportional to how attached-to you already are. Every link you gain raises your odds of the next link — advantage compounds mechanically, like interest. Run the rule long enough and the degrees settle into a power law: on the log–log panel, a straight line of slope near −3, meaning a few enormous hubs above a vast, thin tail. Flip the compare toggle and the same arrivals, choosing at random instead, produce a society of modest equals — the histogram sags into exponential decay, and no giant ever forms. The inequality was never in the nodes. It was in the attachment rule.
What to try
The retrospective illusion
Let a run finish and look at the champion: central, everywhere, indispensable. The story writes itself — obviously the best. Now press rewind. A different node wears the crown, radiating the same inevitability. The hall of winners is a record of monarchs who were each, in their own history, beyond question — and who, in the history next door, never existed.
This is survivorship narration, the human interface of preferential attachment. We meet history exactly once, after the compounding has finished, so we read the output — degree, fame, fortune — as if it were the input. The winner's qualities are real, and afterward they are enumerated as causes; the equally-qualified nodes that arrived at tick 200 are not enumerated at all, because there is nothing to enumerate them for. Prose cannot cure this illusion, because every example prose can give is drawn from the one run that happened. The rewind button is the antidote: it shows you the alternate runs history never lets you see.
What follows — and what doesn't
For individuals: enter early — or enter niches young enough that early is still available. In a mature network the top slots are fossilized; in a new one, a week of luck can compound for a decade. Being tenth-best in a field that is forming can beat being second-best in a field that has formed.
For evaluators: discount accumulated counts as evidence of quality. A citation count, a follower count, a fortune is partly a measurement and partly a fossil of early luck; the model's precise claim is that observed inequality overstates quality differences, by an amount you can watch accumulate above. When you must judge, prefer per-encounter signals — what happens when the thing is actually met — over totals of how often it has been met.
For designers: ranking by popularity is not measuring this dynamic; it is building it. Recommender systems, bestseller lists, "trending now" — each is a preferential-attachment engine bolted onto human attention. The model also exposes the lever: any dose of randomized exposure mixed into the ranking is a dose of uniform attachment, and you have seen what uniform attachment does — it thickens the middle and lets the late-and-good be found.
The mapping
| Mathematics | Life |
|---|---|
| arrival | Each new customer, reader, dollar — deciding, right now, where to attach. |
| attach ∝ degree | Choosing by the crowd, the citation count, the follower number: joining what's joined. |
| the hub | The institution "everyone knows" — known, increasingly, because known. |
| the rewind | History's untaken alternate runs: same rules and same talents, different monarchs. |
| early-mover share | How much of "greatness" is a timestamp — the first 5% of arrivals holding most of the crowns. |
| the fitness dial β | How much quality can fight timing in your domain — an empirical setting, not a slogan. |
Where the metaphor tears
Barabási–Albert is the baseline you compare reality against, not a full account of any real market. Actual domains mix quality, marketing, geography, network topology, node death, and one-off shocks; some "hubs" were built by fiat, not accretion. The fitness dial is the honest gradient between pure luck and pure merit — and its right setting is an empirical question, per domain, dodged equally by the cynic who leaves it at zero and the hagiographer who cranks it to three.
The tail is poor in links, and links are all this model counts. A four-table café with devoted regulars, a paper read deeply by the eleven people it was for, a life deliberately lived off the leaderboard — link-poverty maps onto misery only through a smuggled premise. Some tails are starved; some tails are chosen. The instrument cannot tell them apart, and neither can a follower count.
The model does not say quality is fake — at every β above zero, quality genuinely shifts the odds, and at high β it genuinely wins. Its actual lesson is quantitative and less quotable: observed inequality overstates underlying quality differences by an amount the mechanism itself generates. The interesting work is measuring that overstatement in your domain — not choosing between two comfortable slogans that both excuse you from measuring.