polyAether is a computer program that places small bets on how hot it will get tomorrow in various cities — and tries to be right more often than the going price says it should be.
That one sentence is the whole idea. Everything in this book is just an honest, careful unpacking of it. You do not need to know anything about finance. You do not need to know anything about weather. You do not need to have ever placed a bet in your life. By the end of these ten short chapters, you will understand not only what the program does, but why anyone would expect it to work — and, just as importantly, the specific ways it could fail. We are going to be as honest about the second part as the first.
Let's take that opening sentence apart, piece by piece, in the plainest words possible. There are really only four moving parts hiding inside it: a question that reality will answer, a price that stands for a probability, a forecast that might be sharper than that price, and a set of rules that keep a run of bad luck from ending the game. Get those four, and you have the whole thing.
A bet on tomorrow's high temperature
Every day, in dozens of cities, there is a simple fact waiting to happen: the high temperature — the hottest the air gets during the day, as measured by an official weather station. In New York tomorrow, will the high be above 75 degrees, or below? Nobody knows for certain right now. But it is a clean, checkable question. Tomorrow evening, we will look at the official thermometer reading and know the answer for sure — no debate, no interpretation, no arguing about it later.
That last part matters more than it sounds. polyAether makes small wagers only on questions like that. Not "will it feel nice out," which is vague and unanswerable, but "will the high be above 75," which has a definite yes-or-no answer that everyone can agree on afterward. In practice the markets slice the day even more finely — they don't just ask "above 75 or not," they ask which bucket the high lands in: 71–72°F, 73–74°F, 75–76°F, and so on. Each bucket is its own little yes-or-no question, and exactly one of them will come true.
The program only bets on questions that reality will answer clearly and soon — like which two-degree bucket tomorrow's high temperature lands in, in a specific city. A clear question with a checkable answer is what makes the whole thing possible.
Why a "bet" at all?
Here is the part that surprises people. There are online marketplaces — we'll call them prediction markets — where you can buy a kind of ticket that pays you a fixed amount if some future event happens, and nothing if it doesn't. A ticket that says "New York high will be above 75 tomorrow" might cost, say, 40 cents. If it turns out true, that ticket becomes worth exactly one dollar. If it turns out false, it becomes worth nothing at all. There is no in-between: every ticket ends its life worth either a full dollar or zero.
You can think of the 40-cent price as the crowd's opinion, expressed as a number. When lots of people are buying and selling these tickets, the price gets pushed around by supply and demand until it settles at roughly "how likely the crowd thinks this is." If most people thought the event was nearly certain, they'd happily pay 90 cents for a ticket that pays a dollar; if they thought it was a long shot, nobody would pay more than a few cents. A 40-cent price is the market's way of saying: on balance, about a 40% chance. Chapter 2 builds this idea up carefully from nothing; for now, just hold onto the picture: price is a stand-in for probability.
On a prediction market, the price of a "yes" ticket is roughly the crowd's estimate of the chance the thing happens. A ticket priced at 40 cents is the crowd saying "about a 40% chance." Price and probability are two views of the same thing — a number between 0 and 1 dressed up as a number between 0 and 100 cents.
The whole game in one line
If price is the crowd's guess at the odds, then there is exactly one way to make money here: find the moments when the crowd's guess is wrong, and bet against it. That is the entire strategy in a single sentence. Everything else is detail about how to find those moments, how much to bet, and how not to fool yourself.
Let's make it concrete with a worked example. Suppose a ticket costs 40 cents — the crowd is saying "40% likely" — but polyAether's weather models give it a genuinely better reason to believe the true chance is more like 55%. What is that ticket actually worth? On average, it pays out a dollar 55% of the time and nothing 45% of the time, so its fair value is 0.55 × $1.00 = 55 cents. You're being asked to pay 40 cents for something worth 55. That gap — 15 cents of expected value on a 40-cent ticket — is what we call the edge.
Now watch what the edge does over many bets. Buy that ticket a hundred times in situations exactly like it, and you spend $40. About 55 of them come true and pay a dollar each: $55 back. You are up $15 on $40 risked — even though you were flat wrong on 45 of the 100 individual bets. That is the whole trick: you lose plenty of single bets, sometimes many in a row, but if your probabilities are honestly better than the price, the arithmetic tips your way on average, across many bets. One bet tells you nothing. A thousand bets tell you whether the edge was real.
So the real question becomes: what could possibly let a small program guess tomorrow's temperature better than the crowd? That's where weather comes in.
Weather is guessable — but never certain
Modern weather forecasting is genuinely good. Scientists run enormous simulations of the atmosphere on supercomputers, dividing the planet into a grid and stepping the physics forward hour by hour. polyAether leans on the best of these — a blend of several world-class forecasting systems (with names like GFS, ICON, and ECMWF) combined into one big committee of about 122 slightly-different predictions. Each member of the committee starts from a slightly nudged picture of today's atmosphere and runs its own simulation; where they all land tomorrow is the story. When that whole committee mostly agrees — say, 118 of the 122 members put tomorrow's New York high between 74 and 77 — then "tomorrow's high in this city" is quite predictable, and we can say so with real confidence.
But — and this is the heart of it — weather is never perfectly certain. Chapter 3 explains exactly why: tiny errors in today's measurements grow as the simulation runs forward, so even a perfect model has a fog of uncertainty around it. The useful thing is that a good forecast doesn't just say "the high will be 76." Because we have 122 members voting, it can say "here's the full range of what's likely, and here's how tightly they cluster." When the members are bunched, the forecast is confident; when they're scattered, it isn't. That spread — the shape of the disagreement — is the raw material for a smart bet, because it translates directly into a probability for each temperature bucket.
polyAether's possible edge comes from a large committee of professional weather models. When they agree, it can estimate tomorrow's temperature odds more sharply than a casual crowd — and it only bets when its estimate and the market price clearly disagree. The spread of the committee, not just its average, is what becomes a probability.
The one specific insight
There's a particular pattern this program is built around, and it's worth previewing now because the rest of the book keeps returning to it: crowds tend to overpay for surprises.
People are drawn to long-shot outcomes — the freak heat wave, the surprise cold snap. They'll pay a little too much for the ticket on the dramatic result "just in case," the same way people overpay for lottery tickets and reach for the exciting bet. That extra demand on the tails has to come from somewhere, and it comes out of the boring middle: the everyday, most-likely outcome ends up priced a touch too cheap. Across the markets polyAether studies, this shows up as the crowd pricing uncertainty about 1.3 times wider than the weather models justify — the market's implied "fog" is 30% thicker than the real fog. Chapter 5 is entirely about this. It is, in a sentence, the reason the program exists: not because the program forecasts weather better than the professionals (it doesn't — it borrows their forecasts), but because it turns a well-calibrated forecast into a probability and then finds the buckets the crowd has mispriced.
That word — calibration — is the quiet center of the whole thing, and Chapter 4 is devoted to it. Being "calibrated" means something very specific and testable: of all the times polyAether says "70% chance," almost exactly 70% of them should actually happen. Not the forecast that sounds smartest, but the forecast whose stated confidence matches reality over the long run. The program grades itself on this constantly, using scoring methods (Brier scores, and a calibration check called PIT) that accumulate over hundreds of settled markets. If its "70%" only comes true 55% of the time, the scores catch it, and the edge was never real. Calibration — not speed, not a secret weather model — is where any advantage has to live.
Why "small" bets matter
Notice the word small in our opening sentence. Being right more often than the price implies is only half the job. The other half is not blowing up. Even a genuinely good edge loses individual bets constantly, and a run of bad luck can wipe out someone who bets too big. Go back to the worked example: a 55% bet is a good bet, but it still loses almost half the time, and losing five or six in a row is entirely normal. Someone who staked a quarter of everything on each one could be ruined before the edge ever had a chance to show up.
So polyAether deliberately keeps each bet modest, spreads its bets across many independent cities and days, and has hard limits and an emergency off-switch. It also refuses to bet in situations it can't trust. This turns out to matter enormously in practice: on these weather markets, the resting orders are lopsided. There is real depth on the sell side — market-makers posting large walls of tickets for sale — but the buy side is thin, and many of the model's apparent "edges" are against tiny fractions-of-a-cent orders on extreme tail buckets that no serious money is behind. The program's price and size gates correctly reject those, so on a typical scan only a couple of the roughly eighteen candidate edges actually clear the bar. The honest consequence is that polyAether trades rarely — placing zero trades on a given cycle is frequently the correct, disciplined outcome, not a malfunction. Chapter 8 covers this survival instinct in full.
polyAether is, right now, running strictly on paper — placing pretend bets to test itself, with real money nowhere involved and no proven track record yet. The paper accounting does run end-to-end: a bet opens when a genuine edge clears the gates, the position is held across cycles, it settles on the real official temperature reading, and the resulting profit or loss flows into the tracked balance. But "the plumbing works" is not "the strategy is proven." This book explains the idea and the machinery honestly, including where it might be wrong. It is educational, not a promise of profit and not financial advice.
The journey from here
You now have the whole picture in miniature: a program bets small amounts on tomorrow's city temperatures, using top weather models to turn a well-calibrated forecast into probabilities, spotting the buckets where the market's price disagrees with the real odds — especially the crowd's tendency to overpay for surprises — while betting rarely and carefully to guard against ruin. The remaining chapters simply zoom in on each piece, in the order you'd naturally ask about them.
No step in this book assumes you remember jargon from a previous one — each term is re-anchored where it matters. Turn the page and we'll start with the thing everything else rests on: what a prediction market actually is.