The probability layer of the internet

This series began with the smallest object in finance: a contract that pays $1 if it rains tomorrow and nothing if it stays dry.¹ A toy. But the toy carried a whole idea inside it — that the price of that contract, the number a crowd will pay for it, is the probability of rain. Not an opinion about rain. Not a forecast you have to trust. A live, money-backed estimate of how likely the future is, readable off a screen like a temperature.

Twenty-three pieces is a long way to walk from one contract. We took the walk on purpose, because the destination only makes sense once you've seen every joint in the machine. So before the vision, the map of the road we travelled — and then where the road runs out.

What the journey built

We began on the ground floor: a market is just a place to trade contracts that pay out on whether something happens, and the fair price of a yes/no contract collapses to the crowd's probability — a contract priced as a probability, in plain English. Then we made the claim testable. A price that calls itself 73% is making a frequency promise you can grade with calibration and a Brier score — and liquid markets pass, because the only forecaster that goes broke when it's wrong is the one that stays honest.

From there, the mechanism. How an empty market gets its first quote (an automated market maker with bounded loss), who declares what actually happened (resolution, oracles, disputes), and whether you can trust the number at all (manipulation, insider trading, and the market's habit of correcting itself by paying anyone who shoves the price back toward the truth). That was February — the hard parts, the plumbing that turns a clever idea into something you'd stake money on.

Then the reframing that changes everything: these are not bets, they are derivatives. A binary option on a real-world outcome, collateralised and settled — which means you can hedge with it, offloading a risk no insurer will underwrite. And derivatives have a regulator. In 2020 the U.S. CFTC recognised event contracts as a distinct asset class,² and that single decision is the hinge the whole category swings on — it turned a grey-zone product into a licensed one.

The rest followed the way these things always do once the asset class is legitimate. A business case (the crypto-exchange playbook, run again — license by license, winner-take-most, the first regulated venue keeping most of the liquidity). The wider markets, and their limits (thin order books, long horizons, reflexivity). And the frontiers: markets that score AI models, markets that govern organisations, the perpetual-futures world arriving onshore. Each piece assumed the one before it. Stacked up, they describe not a product but a capability — and a capability wants to become infrastructure.

The thesis: a probability layer

Here is the whole argument in one move. Event contracts are not a category of website where people bet on the news. They are becoming an infrastructure layer — a standard way to take any question with a clean answer and turn it into a live, money-calibrated probability that anything else can consume.

The internet is built in layers, and the useful ones disappear. TCP/IP moves packets; HTTPS secures them; almost nobody thinks about either, and everything runs on top. A layer earns that invisibility by doing one narrow job so reliably that the things above it can simply assume it. The claim of this series is that there is a missing layer of exactly that kind — one that prices the likelihood of real-world outcomes — and prediction markets are how we build it.

What does the layer actually do? It takes the price-equals-probability identity — the rain contract, scaled up — and exposes it as a feed. One job, stated cleanly:

Below the layer: real-world outcomes — elections, rate decisions, shipments, model benchmarks, a port reopening. Inside it: a market, doing the aggregation Hayek described, collapsing thousands of private fragments of knowledge into one public number.³ Above it: a probability feed — an API, a data stream — that news, finance, software, and institutions read the way they already read a stock ticker or a weather forecast. The market is the sensor; the feed is the layer; everything else is an application.

Phrased this way, the question stops being "will people gamble on the news?" and becomes the question every infrastructure layer eventually answers: once a live probability for anything is cheap and standard, what gets built on top that couldn't exist before?

Where it plugs in

The layer isn't a forecast; it's an interface. Four places are already reaching for it.

News. The most visible plug is live. In December, CNN and CNBC both struck deals to put market-implied odds on air — at CNN, the chief data analyst reads them out next to the polls.⁴ That was controversial, and fairly so. But strip the controversy and look at the shape: a newsroom decided that the crowd's price was a more honest live number than the panel of pundits beside it. That is a probability feed being consumed, on television, today.

Finance. A probability is a hedge waiting to be priced. If your business turns on a central-bank decision, a mild winter, or a port reopening on time, an event contract lets you own the other side of that risk instead of merely watching it. And the institutions are treating the output as a data product, not a casino: the Intercontinental Exchange — the owner of the New York Stock Exchange — committed up to $2B to Polymarket to become the global distributor of its event data, a deal that completed in March, piping crowd-implied probabilities alongside its securities feeds.⁵ When the company that runs the NYSE buys the feed, it is telling you what layer it thinks this is.

AI. This one is the most interesting, because it runs both directions. Agents are natural traders — tireless, dispassionate, able to read every source — so they will increasingly set the prices in these markets. And markets are a natural way to score AI: stand up a contract on whether a model will top a benchmark, or whether a claim is true, and you get a money-weighted answer no leaderboard can fake. Machines trading the future, and the future grading the machines — I walked through both in When AI trades the future.

Governance. The most radical plug is the oldest idea in the series, taken to its conclusion: don't just read the probability, decide on it. Futarchy — "vote on values, bet on beliefs" — uses one market to price whether a policy will hit an agreed goal, and lets that price drive the choice. It's unproven at scale and I'm not selling it. But it's the purest expression of the thesis: a probability layer isn't only something you watch, it's something you can wire an institution to.

Why now

Layers don't appear because someone draws a nice diagram. They appear when the conditions underneath them flip from impossible to inevitable. Four flipped, roughly at once.

The regulatory unlock. For thirty years this was an academic curiosity with no legal home. The CFTC's 2020 recognition of event contracts as a regulated asset class is the bedrock — without a licensed venue there is no institutional feed, no NYSE owner writing a cheque, no newsroom comfortable putting the odds on screen.

The business proof. A layer needs someone running the rails at scale, and the unit economics now exist. Kalshi's revenue went from $1.8M to $24M to $263M across 2023, 2024, and 2025 — one of the most compressed revenue ramps in exchange history — and in March it was valued around $22B.⁶ Category volume tells the same story: under $100M a month in early 2024, and roughly $24B a month by April 2026.⁷ That is not a fad's shape.

Institutional buy-in. The tell isn't that money arrived; it's what it bought. ICE didn't buy a casino — it bought the data, the right to distribute the probabilities. That is a vote for the layer, made with two billion dollars.

Expansion past event contracts. And the rails aren't staying narrow. On May 29, the CFTC cleared the way for regulated crypto perpetual futures to come onshore — Kalshi listed a Bitcoin perp the same week.⁸ Same regulator, same licensed-venue logic, a far larger surface. The infrastructure being laid for event contracts is general-purpose, and it is being extended in real time.

Where the two threads converge

I write about a second thing on this blog: AI data for the physical world — the argument that the next decade of AI is gated less on model size than on trustworthy data about what is actually happening out there. It looks like a different obsession. It isn't.

An internet flooded with synthetic everything has a new scarcity: a reliable answer to is this true, and how likely is that? A model can generate a thousand confident claims a second; what it cannot manufacture is a number that costs someone money to be wrong about. That is precisely what the probability layer produces. In a world where text is free and truth is expensive, a market is one of the few mechanisms that prices truth instead of asserting it — because it makes confidence pay for itself. The two threads of this blog are the same thread: how do we keep a public, accountable signal about reality when reality gets cheap to fake.

The honest edge

This series earned whatever credibility it has by refusing to oversell, and I'm not going to break that on the last page. The probability layer is a direction. It is not, today, a destination — and the gap is real.

Most markets are thin. A probability is only as good as the capital standing behind it, and outside a handful of headline questions the order books are shallow, the spreads are wide, and the "live number" is mostly noise. Reflexivity bites: when a market's own price starts steering the outcome it's supposed to measure, the sensor and the thing it senses get tangled. And the honest headline is the one this series has repeated since February — roughly 80% of today's volume is sports.⁹ The probability layer, as actually traded right now, is mostly a very good way to price football games.

So hold both truths at once. The thing people do with these markets today is overwhelmingly entertainment. The thing the rails can become — a money-calibrated probability for anything with a clean answer, distributed like a data feed and built upon — is generational. Sports is the on-ramp, not the road; it funds the liquidity and the licenses while the forecasting and data layer is what the institutions are quietly buying toward. Anyone who tells you the layer is already here is selling. Anyone who tells you it isn't coming hasn't looked at who's laying the track.

That bet is the one I get up for. Two of the three hard problems in this series — bootstrapping liquidity, and resolving the truth — are engineering. The third, trust, is mostly regulatory: a market that pays out on real-world events is a licensed financial product, granted one jurisdiction at a time, and the venue a market trusts is usually the one that showed up early and won the license to operate inside the rules rather than around them. That is Seeker's bet: win the licensed venue in each market. The MVP is live; the license is the goal, not a claim I get to make yet.¹⁰

But step back from any one company, because the through-line of all twenty-four pieces was never a company. It was a single, almost suspiciously simple sentence: a price is a probability. Follow it far enough and you get a society that can read its own best guess about the future — priced in the open, updated in real time, with money keeping everyone honest — instead of arguing over stale polls and confident pundits. We started with a contract that pays a dollar if it rains. The layer is what you get when you let that one honest number price everything that has a clean answer. Most of it is still sports. The rails are generational. And the next number on the screen — 73% — you now know exactly what it means, and exactly how you'd check it.