Resolution Risk: What 373 Polymarket Rulebooks and 6,495 Kalshi Contracts Actually Say

Every prediction-market trader eventually asks the same question in some form: how do I know this market pays out the way I think it does?

It is a harder question than it sounds. You can be completely right about the event and still lose, because the contract was grading something narrower, or slower, or measured by a source you never looked up. Traders call this resolution risk, and almost everyone agrees it matters. Almost nobody quantifies it.

So we did. On July 29, 2026 we pulled the live resolution rules for every open market we could reach on both major venues and read what the text actually commits to. The sample:

Four findings came out of it. Together they explain why "the same event" can pay differently on two venues, and they turn resolution risk from a vague worry into something you can check in about ninety seconds.

Finding 1: most Polymarket rulebooks do not name a source you can go check

The single most useful thing a rulebook can do is point at a specific, public, checkable source. If the rules say "according to the Federal Reserve's official statement," you know exactly what to watch and exactly what would settle an argument.

That is the minority case.

The second number is the one worth sitting with. A majority of these markets delegate the decision to an aggregate judgement — "a consensus of credible sources" — rather than to a document anyone can pull up. That is not necessarily wrong. For a lot of events there genuinely is no single authority. But it means the honest description of what you are trading is: the event, as later summarised by an unspecified group of reporters, as later read by a proposer.

And it is not the small markets doing this. Sorted by volume, the top of the book looks like this:

Notice the pattern. The Fed market — the one with an obvious official source — is the one that links it. The geopolitical markets, where "invade" and "acquire" are exactly the words people would fight about, are the ones that fall back on consensus reporting.

There is a second-order problem here too. Polymarket's API exposes a structured resolutionSource field, which sounds like it should let you filter for this automatically. In our sample, 1,175 of 1,199 markets left it empty. The source, when it exists at all, lives in prose inside the description. You cannot screen for it. You have to read.

Finding 2: on Kalshi, the sentence that decides your payout is usually not the one you read

Kalshi structures rules differently, and the difference matters more than it first appears.

Each contract has a rules_primary — a short trigger sentence — and a rules_secondary, which is where definitions live. The primary is what reads like the contract. The secondary is what functions like the contract.

Here is a live example from the "Who will be the next Secretary General of NATO?" event. The primary rule reads:

If [candidate] formally holds the position of Secretary General of
NATO, and is the first such subject to do so after Issuance, then
the market resolves to Yes.

Clean, readable, and almost entirely dependent on two words: formally holds. The secondary rule then defines them — and the definition explicitly includes assuming the role "by appointment, election, succession, confirmation, designation, or acting/interim assignment."

That last clause changes the trade. A caretaker appointed for six weeks resolves the market. If you priced "who gets the permanent job," you priced a different contract than the one you bought — and nothing in the primary rule told you so.

The source asymmetry is just as stark. Something that reads like a settlement source appears in the secondary rules of 10.4% of contracts, and in the primary rules of 0.5%. If you are reading only the sentence the interface leads with, you are reading the part that is least likely to tell you where the answer comes from.

Finding 3: rulebook length tracks contested definitions, and it is free information

The median Polymarket rulebook is 133 words. But 21% run past 200 words, and the distribution has a long, revealing tail:

Every one of them is a market about whether a contested geopolitical state of affairs has "really" happened. The length is not bureaucratic padding — it is the market maker pre-litigating an argument they know is coming. An 800-word rulebook is a signed confession that the plain-English version of the question does not survive contact with reality.

We expected rules length to scale with market size, and it does not. Splitting the sample into volume thirds, median rules length runs 144 / 131 / 128 words top to bottom, and the share with a linked source barely moves (35% / 35% / 37%). Big markets are not better documented than small ones. Volume does not protect you.

Some other language patterns worth knowing, across the 373 rulebooks:

• 26% contain narrowing language — "only if," "solely," "must" — a clause that shrinks the market relative to its title.
• 10.5% contain an explicit exclusion clause ("will not count," "will be ignored").
• 5.4% specify a 50-50 backstop for genuinely undecidable cases. The other 94.6% do not tell you upfront what happens if the question turns out to be unanswerable.

Finding 4: the correction mechanism has a price and a clock — and they differ per market

Even a badly-worded market is survivable if a wrong answer can be challenged. Both venues have that mechanism. Neither is uniform.

Polymarket resolves through UMA's optimistic oracle. Someone proposes an outcome with a bond attached; if nobody disputes it within the liveness window, it becomes final. Polymarket's documentation describes a bond of roughly $750 and a two-hour challenge period, and its own docs state the point plainly: "The market title describes the question, but the rules define how it resolves."

But the per-market parameters vary more than the docs suggest. In our sample:

A $50,000 bond makes a market very expensive to grief — and also very expensive to defend. If you hold $300 of a contract whose dispute bond is $50,000, you are not the one who is going to challenge a bad proposal. You are relying entirely on someone larger noticing and caring on your behalf, inside a two-hour window.

And the window is not always two hours. 23 markets in the sample carried a shortened custom liveness — 10, 15, or 30 minutes — concentrated in fast-moving crypto and sports markets. Those are precisely the markets where an overnight proposal is most likely to go unwatched.

Kalshi works differently. As a CFTC-regulated exchange it settles its own markets against a named source agency, and its rulebook defines a "Source Agency" as the agency that publishes the underlying or expiration value for a contract. There is no permissionless dispute auction; there is an exchange process. In our sample, every contract was flagged can_close_early, and settlement timers clustered at 30 minutes (5,219 contracts), 5 minutes (855), one hour (197), and 14 seconds (166).

These are two genuinely different risk profiles, not two versions of the same one:

Why the same event can pay differently on two venues

Put the findings together and the answer stops being mysterious. Two contracts on "the same" event can differ on four independent axes:

1. The source. One names an agency; the other defers to consensus reporting.
2. The definition. One counts acting/interim appointments; the other says "formally."
3. The decider. A bonded proposer versus an exchange settlement desk.
4. The clock. A two-hour permissionless window versus an internal review with no public deadline.

Any one of those can flip a payout while the underlying event stays identical. This is why a price gap between two venues quoting the "same" question is not automatically an arbitrage. Sometimes the gap is the market correctly pricing two different contracts. The only way to tell is to read both rulebooks and check whether they are actually grading the same thing.

A 90-second pre-trade checklist

None of this requires a spreadsheet. Before you take a position, run these six questions against the rules text:

1. Which sentence names the decider?

Find it. If there is no named source — if it is "credible reporting" — you have found a real risk factor, not an absence of one.

2. What is the definition of the key verb?

On Kalshi, open the secondary rules. On Polymarket, read past the first paragraph. The market usually turns on one word.

3. What is the deadline, and in which timezone?

89% of Polymarket rulebooks specify ET and 79% give an explicit clock time. When a market does not, that vagueness is doing work.

4. Is there an exclusion clause?

Search for "not count," "excluded," "will be ignored." One in ten markets carves something out that the title implies is included.

5. What happens if the question is unanswerable?

Only about 5% state a 50-50 backstop upfront. For the rest, an ambiguous world state is an open question, not a defined outcome.

6. What does it cost to challenge a wrong answer, and how long do you have?

Check the bond and the liveness window. If the bond is larger than your position by two orders of magnitude, your protection is other people's diligence.

What this research does not show

Worth being explicit about the limits, because the numbers above are easy to over-read.

This is a snapshot of open markets, so it says nothing about how often ambiguous rules actually produced a contested settlement — that would need resolved-market data and dispute records. The Polymarket sample is ordered by volume and therefore skews large. The Kalshi sample sits inside open events and skews heavily toward auto-generated sports and series contracts, which are more templated than the exchange's headline markets. And the classification is pattern-matching over text: a rulebook that names a source in prose without linking it counts as "unlinked" here, which undercounts the careful ones.

What it does show is a structural fact that holds across thousands of live contracts: the title is the marketing, and the resolution text is the contract — and on both venues, the resolution text is stored somewhere the interface does not lead you.

Making this a habit rather than a chore

The reason most people skip this is not laziness. It is that the rules, the source, the deadline, the related market on the other venue, and the news that would actually trigger the clause all live in five different places. Checking them properly takes longer than the trade is worth, so it does not get done.

That gap is the thing worth closing, with or without a tool. Catalyst was built around the same habit this research points to: when a market moves, pull the rules, the source, the timing, and the surrounding evidence into one view instead of five tabs, so that "what would actually have to be true here?" is a question you can answer in the moment rather than after settlement.

But the discipline stands on its own. Read the rules before the chart. Find the source before the entry. And when a market's rulebook runs 800 words, believe it — someone already knew that question was going to be a fight.

Source notes

• Data. Polymarket Gamma API (/markets, open + active, ordered by volume) and Kalshi trade API (/events with nested markets, open status). Both pulled July 29, 2026. Figures in this article come from that snapshot.
• Polymarket resolution mechanics, UMA optimistic oracle, bond and two-hour challenge period, and the "the rules define how it resolves" wording:
  https://docs.polymarket.com/concepts/resolution
• UMA optimistic oracle liveness and dispute escalation to the DVM:
  https://docs.uma.xyz/faqs
• Kalshi settlement, source agencies, and the role of per-market rules:
  https://help.kalshi.com/en/articles/13823821-market-faqs
• KalshiEX LLC Rulebook (definitions of "Source Agency," "Underlying," and "Expiration Value"), as filed with the CFTC:
  https://www.cftc.gov/sites/default/files/filings/orgrules/25/07/rules07012525155.pdf
• Percentages are regex classifications over live rules text and are approximate. Individual contracts should always be read directly.