Verified Platforms
Quick Links
Where to Stay Secure
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
The probability on the screen is not the thing you are betting on.
The sentence underneath it is.
A prediction market quotes you a number. 73 percent yes, 27 percent no. That number is what everyone watches. It is not what decides whether you get paid. What decides that is a clause you probably scrolled past, the one that says how the event will be judged, who judges it, and when the judging stops. Prediction market resolution risk is the gap between the event you think you bet on and what that clause actually says.
This is a Kodex walkthrough with Ava, the architect who reads structure before price. You will work through one real contract the way she does, clause first and number last, until you can open any market and find the risk the odds are hiding.
Ava does not start with the chart. She scrolls past the 73 and stops on the fine print.
"This line is the product," she says. "The rest is decoration."
Strip a prediction market down and one sentence is left. It names a source of truth, sets a condition, and fixes a cutoff. "Will this team win the World Cup, per this governing body, by this date." The price, the volume, the green and the red all hang off that sentence.
Change one clause and you change the bet. "Win" can include a penalty shootout or exclude it. "By this date" can count a result that lands a minute late or throw it out. The wording does the deciding, and it is set before you arrive.
You are not buying the outcome. You are buying the platform's promise to read that sentence a particular way.
That is a different object than the event itself, the same way who controls a token defines what you hold, not the asset it points at. The World Cup happens in the world. Whether your contract pays depends on a clause sitting on an exchange's server and on the people and code that interpret it.
Usually the two line up. A team wins, the clause reads the obvious way, the market pays. The risk lives in the cases where they come apart, where the event turns messy and a sentence written in advance has to be applied to a situation nobody pictured. That is the moment you learn whether you were reading the structure, not the headline, or whether you never read it at all.
Three parts of the sentence carry almost all the risk.
The source of truth. Who or what officially says the event happened: a named body, a specific data feed, a website checked at a stated time. The choice is not cosmetic. The same election can be called by one wire service hours before another, and a contract bound to the slower one stays open while the faster one has already paid. The fragility of a prediction market lives in the data that resolves it, not in the trade you placed. If the named source is slow, offline, or hair-splitting, the real world can be settled while your position is not.
The timing. When exactly does the event count, and when does the window shut? A goal in stoppage time, a result overturned on review an hour later, a delay that pushes the outcome past the cutoff. The clock in the clause is rarely the clock in your head.
The edge cases. This is where it gets expensive. Does the market read the question in its spirit or to the strict letter? The two can settle the same event in opposite directions.
One phrase can flip yes to no.
It has happened in public. A halftime-show market on Cardi B settled in contradictory directions across different platforms, on the same performance, because each contract defined "performed" a little differently. Years before that, on Augur, token holders argued for weeks over what a single market's wording meant before anyone was paid. Same event in the world. Different sentences. Different winners.
Ava is not rattled by this. She expects it.
"Ambiguity is not a defect they forgot to fix," she says. "It is the part of the product you are underwriting."
So a contract is disputed. The event was messy, the wording strains, and your side looks right by any common-sense reading. Who decides, and what is your recourse?
The honest answer is that you have almost none, and the shape of it depends entirely on the venue.
On Kalshi, a centralized exchange, a markets team determines outcomes and an Outcome Review Committee issues binding determinations when a result is contested. On Polymarket, resolution runs through the UMA Optimistic Oracle, and a dispute escalates to UMA's Data Verification Mechanism, where token holders vote the answer. DeFiRate has documented seven Kalshi resolution errors and four Polymarket controversies, which tells you a contested settlement is a routine event, not a freak one.
Ava sets the two machines side by side before she touches either.
"Same question for both," she says. "When the sentence and the world disagree, whose reading wins. Never yours."
One venue carries that risk as a centralized desk's discretion, the other as a token-holder vote, and the symmetry is the part worth seeing. A desk can read its own clause the way it chooses to defend. A vote can be swung by whoever shows up holding enough tokens. Neither path keeps a seat open for you.
| Kalshi | Polymarket | |
|---|---|---|
| Who resolves it | Markets team, escalating to an Outcome Review Committee | UMA Optimistic Oracle, escalating to the Data Verification Mechanism |
| Dispute path | Internal review, binding committee determination | On-chain dispute, token-holder vote |
| Your recourse | None binding; you accept the committee's call | None binding; you accept the oracle's call |
| Where the wording risk sits | The exchange's reading of its own clause | The token vote's reading of the clause |
Read across either column and the same fact lands. You are short the platform's interpretation of its own sentence. If the resolver reads the clause against you, real-world facts do not buy you anything. You were not wrong about the world. You were wrong about the wording, and the wording is what pays.
That is the cost no probability on the screen shows you.
In June 2026 the asymmetry got sharper, and it happened quietly.
Kalshi revealed an internal AI agent named Harrison, built on Anthropic's Claude, that it uses to stress-test event contracts before they list. Harrison cross-references each draft against a library of more than 500 vetted templates, runs simulations for structural loopholes and semantic ambiguities, flags problem phrasing, and compresses a review people used to do by hand. Before Harrison, Kalshi hired Yale debate champions to attack its own contract wording. Now a model does it at the speed of listing.
It reaches into resolution too. When an event closes, one markets-team member enters the outcome, a second enters theirs independently, and the agent checks that the two humans agree with each other and with its own reading before the contract settles.
Three reads, one sentence.
It is easy to take that as reassuring. Tighter contracts, fewer disputes, a cleaner book. Ava reads it the other way.
"Ask who the hardening is for," she says.
Sharper wording closes the gaps a careful reader used to find, the small ambiguities where you could take the other side of a sloppy sentence. Close them and the exchange writes a cleaner contract. It also writes one that reads the way the house intended, because the house pressure-tested every word before you saw it. The edge that rewarded reading closely gets industrialized away.
This is running at scale right now. Kalshi cleared a record near $18 billion in notional volume in May, took roughly $5.1 billion in its opening World Cup week, and the tournament has been live since June 11. The contracts in front of you this week are among the first written by a system built to leave nothing exploitable in the wording.
Harrison does not run the exchange.
It writes the sentences you are betting against.
No. It makes them safer for Kalshi.
The distinction is the whole point. Harrison lowers the exchange's operational and dispute risk: fewer contested settlements, fewer reputational fires, faster listings. Your risk is the wording reading against you, and a contract hardened against ambiguity is one where the intended reading is locked in tighter than before. Their exposure falls. Yours does not move, and the edge you had to push back against it gets smaller.
So price it. The way the Survival Framework treats any cost you cannot control, fold resolution ambiguity into the number before you size anything. A clean, binary, well-sourced question carries little of it. A fuzzy event with a vague source and a soft cutoff carries a lot, and the quoted odds will not subtract it for you. If you cannot tell which one you are looking at, you are not pricing the bet.
You are guessing at it.
Here is what Ava actually does before she risks anything. Four questions, in order, and the price is the last thing she looks at.
Run those four and you are doing for a contract what Pattern Intelligence does for your own trading: reading the structural exposure instead of the surface. The number tells you what the crowd expects. The clause tells you what has to be true for you to get paid.
Those are not the same thing, and only one of them is enforceable.
A few questions come up the first time someone runs that checklist.
It is the risk that a contract settles on how its wording is interpreted rather than on what actually happened. The event can go your way in the world while the contract resolves against you, because the clause defined the outcome differently than you assumed.
Not in the sense of secretly fixed results. The structural reality is quieter: you hold the platform's interpretation of its own contract, and when the wording is ambiguous, the resolver's reading is binding and final. The risk sits in who writes and judges the sentence, not in a hidden thumb on the scale.
Harrison is an internal AI agent, built on Anthropic's Claude, that Kalshi uses to stress-test event contracts before listing and to cross-check outcomes at settlement. It checks drafts against more than 500 vetted templates and flags ambiguous phrasing. It reduces the exchange's dispute risk. It does not reduce your wording risk.
Yes, when the wording and the event diverge. If the named source is slow, the cutoff lands awkwardly, or an edge case reads strictly against the obvious outcome, a contract can pay the side that looks wrong by common sense and right by the letter.
Read the resolution clause first. Identify the source of truth, the exact timing and cutoff, the edge cases under a strict reading, and who holds the binding final say. Only once those are clear should the quoted probability mean anything to you.
The number is the last thing to read, not the first.
A prediction market hands you a clean probability and asks you to trust it. The discipline is to distrust it just long enough to read the sentence it rests on. Find the source, the clock, and the edge cases, then find out whose reading is final. Only then look at the price, and decide whether it pays you enough to carry the one piece of the trade you will never control: the wording.
