From the Cheap Seats

on telling a conspiracy from a coincidence of incentives, at a distance

There is a town below with eight gas stations and, possibly, a secret. Skip to it ↓

You are driving home on a Tuesday evening and the gas station on the corner has jumped thirty cents since lunch. So has the one across from it. So has the one diagonal to that, and by the time you pass the fourth corner you are not entertaining hypotheses anymore, you are entertaining exactly one: somebody made a phone call. Four businesses do not leap the same amount, the same direction, the same afternoon, by accident. You know collusion when you see it. Everyone knows collusion when they see it.

I want to take that certainty apart tonight — not because it’s foolish, but because it’s doing something interesting. It is reading intention off of pattern. And pattern, on this site, is the one thing we’ve learned to stop trusting as a signature of intention. A flock turns as one bird with no leader; a termite cathedral goes up with no architect. The question this page asks is narrower and more uncomfortable than either of those, though. It isn’t “can order arise without a plan?” You’ve seen that a dozen times by now. It’s: can you, standing outside, tell the difference between the order that arose and the order that was arranged? Because both exist. Real price-fixing cartels are real, they get prosecuted, and they steal real money. And the view from the cheap seats — the only seats most of us ever get — shows the same show either way.

The dance

Here is what gasoline prices actually do in a lot of cities, and it is stranger than collusion. The price doesn’t sit still and it doesn’t follow the oil market tick for tick. It saws. It falls a cent or two a day for a week or three — a long, grinding slide — then leaps twenty or thirty cents in a day or two, every station together, and starts the slide again. Plotted, it looks like a row of shark fins. Economists call the shape an Edgeworth cycle, after a Victorian economist who saw it coming a century early, and Eric Maskin and Jean Tirole worked out the modern theory in 1988: two firms undercutting each other a penny at a time, all the way down to cost, until someone cracks and jumps back to the top, and the other follows, forever.

The mechanism needs nothing but self-interest and eyesight. Undercut the corner by a cent and tonight the whole town’s traffic is yours — so everyone undercuts, a cent at a time, and the price grinds toward wholesale cost. At the bottom nobody is making money on gas, and everyone waits, because whoever raises first eats a lonely, expensive day of being the most expensive sign in town. That’s a game of chicken, and eventually somebody — the hungriest, the boldest, the one whose franchise agreement is due — blinks, and posts a price thirty cents up. For a day he looks like a fool. Then the station across the street realizes relief has been offered, matches him, and inside of two or three days the whole block has leapt. From your car window: a coordinated thirty-cent jump, four corners, one afternoon. From inside: eight strangers who never spoke, each solving its own little problem, using the only channel any of them ever needed — the number on the sign.

This is not a theorist’s fable. Retail gasoline saws this way in Toronto, in the American Midwest, and most famously across Australia’s big cities, where the cycles run so regular and so documented that the national competition regulator — the ACCC, the same agency whose job is catching cartels — publishes which day of the week is cheapest to buy gas in each capital, the way a tide table publishes the tides. Their conclusion, after formally investigating the very pattern that convinces every commuter of a conspiracy: the cycles are what competition looks like in this market. The jump you saw wasn’t the handshake. The jump was the chicken game ending, on schedule, the way it ends every few weeks.

The price sign is the whole conspiracy: a message every rival can read, that no one had to send.

Notice what the sign is doing in that story. The ants of The Trail Is the Plan coordinate through scent laid on the ground — marks left in a shared world, read by strangers. A posted fuel price is exactly that: the most public pheromone in town, updated daily, legible at forty miles an hour. When people say “the stations didn’t need to call each other,” this is why. The communication is already built into the pavement.

Four corners, no meeting

Like the flock’s three rules, the whole dance fits in a pocket. Each station, each morning, runs something like:

• If anyone matched your price or beat it, undercut the corner by a cent or two — unless that would mean selling gas at a loss.
• If the margin is gone, hold at the floor and wait — someone will crack before you do. Unless they don’t. Occasionally, blink first.
• If somebody has broken for the surface, follow them up — provided you’re hungry enough to want the relief; a comfortable station lets a lone fool stand alone. And a lone fool, unfollowed, slinks back down.

That last clause matters more than it looks. In the simulation below, a station sometimes misjudges — jumps too early, when the others are still fat enough to enjoy watching him dangle — and the restoration fails. He stands up there for a few days, most expensive sign in town, then gives up and comes back down. Real cycling markets show these failed restorations too, and I’d ask you to notice what they are: the fingerprint of nobody-in-charge. A committee that has agreed on Thursday does not produce volunteers who leap on Monday and die on the hill alone. Keep that thought; the statistician downstairs is going to need it.

So: below is a town of eight stations running those three reflexes and nothing else — no channel between them but the signs. Except when there is one. The Open the phone lines button hands them an actual cartel — a genuine, prosecutable agreement to coordinate the jumps — and Deal a new town makes it a game: the town comes up with the phone lines secretly on or secretly off, fifty-fifty, and you get to sit in the cheap seats and call it. A statistician plays alongside you. Watch as long as you like before you guess. That’s the whole page, really: how long do you have to watch, and what do you have to measure, before you honestly know?

Things to try:

The court that couldn’t tell

If it comforts you to know that your Tuesday-evening certainty has also gripped some of the best legal minds of the last century — it gripped them, and then it embarrassed them. In 1939 the Supreme Court looked at eight film distributors who had all, in the same season, adopted the same peculiar restrictions on Texas movie theaters, after each received a letter that named all eight of them on its face. Nobody could produce a meeting. The Court said, in effect: they each read the invitation, saw that it named the others, and complied knowing compliance only made sense if the others complied too — that’s enough. Interstate Circuit became the case that let parallel conduct plus circumstance prove a conspiracy nobody witnessed.

Fifteen years later the pendulum swung back. A Baltimore theater owner sued the film studios because every one of them had refused him first-run pictures — identical behavior, eight for eight, surely a plot. The Court disagreed, and Justice Clark wrote the sentence this page orbits: “this Court has never held that proof of parallel business behavior conclusively establishes agreement… ‘conscious parallelism’ has not yet read conspiracy out of the Sherman Act entirely.” Each studio had its own private, sufficient reason to say no — a suburban theater competing against downtown first-runs was simply a worse deal, for everyone. Eight identical decisions, zero phone calls required. The doctrine that grew from these two cases is the law’s version of this site’s whole thesis: parallel movement alone proves nothing, because parallel movement is what shared incentives produce by default; to convict, you need plus factors — evidence of acts that make no sense unless there was an agreement. The courts, in other words, were forced to invent a third box: not coincidence, not conspiracy — structure. It took them decades, and they had subpoena power. You’re doing it from a moving car.

The phone records

Here is the part that might actually unsettle you, though. In the sim, “Reveal the phone records” ends the argument — records or no records, case closed. Real life has run that exact experiment, and it came out messier. In Geelong, Australia, in the early 2000s, the regulator went after a ring of petrol retailers with what looked like the dream file: telephone records showing station operators calling each other, discussing coming price rises — timing and size — followed, again and again, by the town’s prices marching up together. In 2007 the Federal Court threw the case out. Talking, followed by synchronized pricing, was not enough: the judge found no commitment — no obligation anyone had taken on, nothing anyone was bound to do — and without commitment there was no “arrangement or understanding” under the law. The calls happened. The prices moved. Case dismissed.

Sit with how strange that is. From the cheap seats, an innocent Edgeworth cycle looks exactly like a cartel. And from the best seats in the house — the ones with the phone logs on the table — the line between “we talked, then we each did what we were always going to do” and “we agreed” can still be too fine for a federal court to find. The boundary between emergent coordination and real coordination isn’t just hard to see from a distance. Up close, in the hardest cases, it is genuinely, legally blurry — a line the law has to draw through fog, because the two kinds of order shade into each other with no seam.

The algorithm in the room

The story is not done evolving, because the stations have started outsourcing their reflexes. In 2015 the Justice Department brought its first e-commerce price-fixing case against a man named David Topkins, who sold posters on Amazon and had agreed with rival sellers to run their pricing algorithms in concert — an old-fashioned handshake, executed in code. That one was easy; the handshake was still there. The hard case arrived a decade later: RealPage, whose software set rents for millions of apartments by pooling landlords’ own private numbers and recommending prices back to each of them. No landlord called another landlord. They all just hired the same very well-informed consultant. The Justice Department sued in 2024; in late 2025 RealPage settled — admitting no liability, agreeing among other things to stop feeding real-time nonpublic competitor data into its recommendations. Notice what the settlement regulates: not whether the landlords may watch each other — they may, everyone may — but how fresh and how private the shared signal is allowed to be. The law is groping for the exact boundary this page’s toggle crosses: at what point does everyone-reading-the-same-sign become everyone-on-the-same-phone-call? There may not be a clean answer. The sign and the phone call were always the same kind of thing — a shared channel — differing in privacy, speed, and intent.

What the cheap seats can know

So what are you entitled to conclude, next Tuesday, at the fourth corner? Some honest bookkeeping. The lockstep itself — the thing that felt like proof — carries almost no information: shared costs, shared eyesight, and shared incentives produce it by default, and a regulator that studied the shark fins for years filed them under competition, working. The real tells are quieter and stranger than the jump: the ragged column of stragglers, the occasional volunteer dying alone on the hill. In the model, an honest statistician reading those tells with a season of data almost never accuses an innocent town — and catches the lazy cartel every time. But the same statistician, against a cartel careful enough to fake its lags, catches nothing, forever. Statistics can convict the sloppy; it cannot convict the careful, and it cannot fully acquit anybody. That asymmetry is why cartel cases are made with wiretaps, immunity deals, and the prosecutor’s standing offer — the first conspirator through the door gets amnesty — rather than with price charts.

And the cartels are there to catch. It would be a comfortable ending to tell you the hidden hand is always an illusion; it isn’t. The Geelong operators really did make those calls. Topkins really did rig the algorithms. And the wild-grown sawtooth earns no verdict of innocence either — tacit coordination can overcharge a town with nobody to indict, and what a market reliably falls into is a fact about its incentives, not a seal of approval. What the universe wants — in this market, with these incentives — is the sawtooth, and it will grind one out with or without villains; but the same machinery that makes the innocent look guilty gives the actually-guilty a place to stand where they look like everybody else. Emergence isn’t the alternative to conspiracy. Emergence is the cover conspiracy hides in, one cent wide, six dollars a year, indistinguishable at highway speed.

So the Tuesday certainty gets replaced with something slower and better. What you saw was real: the whole town moved together. What you inferred — the phone call — was one explanation of three, and the boring structural one carries the day more often than either of the exciting ones. Coincidence is the first box, conspiracy the second. The third box is the one this site keeps handing you: structure — strangers, incentives, and a number on a sign that everyone can read. It isn’t the whole answer. It’s the box to check first, before you reach for the one with the villains in it.