Kalshi Uses a Rulebook, Polymarket Uses a Vote, World Uses a Machine: Why Prediction-Market Resolution Is Now the Whole Game

Kalshi Uses a Rulebook, Polymarket Uses a Vote, World Uses a Machine: Why Prediction-Market Resolution Is Now the Whole Game

Every argument about prediction markets in 2026 has been about who is allowed to run one. New York suing for billions, a dozen state gaming regulators issuing orders, California tribes litigating at the Ninth Circuit, a bill to ban sports contracts outright. Almost none of it touches the question that actually decides whether these platforms work: how does the market know who won?

Until this month there were two answers shipping in production. Kalshi settles bets with a rulebook, applied by a licensed operator with a federal regulator behind it. Polymarket settles them with an optimistic oracle, where an outcome is proposed, challenged, and if contested, voted on by holders of a governance token. On September 9, World shipped a third: Chainlink data feeds settle the contract automatically the moment the game ends. No panel, no vote, no dispute window, nobody to appeal to, because the settlement is a program that has already run.

World opened its standalone site at world.xyz to a waitlist exceeding one million users, after operating inside the Phantom wallet since the summer. More than 150,000 markets have been created across sports, crypto, politics, finance, economics and culture, covering every NFL regular-season game, seven soccer leagues, Formula 1, the 2026 midterms and Federal Reserve policy decisions. The site went offline on launch day under the traffic before returning.

Three models, one tradeoff

The three resolution mechanisms differ in every visible detail, but they share a single underlying principle, and it is worth stating plainly: every mechanism that removes discretion also removes correction.

Kalshi’s rulebook resolution means contracts settle according to criteria published in advance, applied by an exchange holding designated contract market status. A named entity makes the determination, a federal regulator supervises it, and participants have a complaints path. The operator can correct errors, and the operator can also make a discretionary call you dislike. Those are the same property viewed from two angles.

Polymarket’s optimistic oracle means nobody can unilaterally decide an outcome, and nobody can correct one after finality. A challenge window opens, disputes escalate to a governance-token vote, and the result finalizes on chain. Those, too, are the same property from two angles.

World removes the most discretion of the three and therefore removes the most correction. Chainlink Data Streams supply the market data, and the Chainlink Runtime Environment executes settlement once a game concludes or a defined event reaches its deadline. The tradeoff is consistent across all three, and pretending one is simply better misses the point.

Where automated settlement clearly wins

Inside a specific range, World’s design is plainly superior. A football match ends with a score. A Bitcoin price at a stated timestamp is a number. A Fed decision is a published figure. For contracts resolving on unambiguous, machine-readable data, a dispute window is pure latency, delaying payout to accommodate an argument nobody will make. Automated resolution pays immediately.

It also removes the failure mode that has damaged the category most. Contested oracle resolutions on token-vote venues have produced the worst reputational damage prediction markets have suffered, because a market resolving against what most observers believed happened is the single thing that destroys confidence in a forecasting instrument. For unambiguous data, the automated design eliminates that possibility entirely.

And it scales. A resolution process requiring human attention constrains how many markets can exist. The 150,000-market figure would be operationally impossible under rulebook administration. Notably, Polymarket itself has moved toward oracle-based settlement for price-based markets, which is confirmation the design is correct for that category.

Where it breaks

The limits sit exactly in the markets people care about most. A contract on whether an official resigns, whether a conflict qualifies as a ceasefire, or whether a statement constitutes an endorsement cannot be settled by a price feed, because the disputed element is the definition, not the measurement. Kalshi’s rulebook answers these in advance, badly or well, and a person applies it. Polymarket’s oracle answers them through challenge and vote, slowly and sometimes contentiously. An automated feed cannot answer them at all.

Data feed failure has no remedy either. If a feed reports incorrectly or late, the system settles on it, and with no dispute window there is no stage at which anyone can say the input was wrong before money moves. A postponed match or an abandoned race is handled by whatever the contract specified in advance, and contracts cannot anticipate everything. On Kalshi you can complain to an exchange and a regulator. On Polymarket you can, in principle, dispute. On World, the settlement is the output of a program that already ran.

The unexamined regulatory question

World lists contracts on NFL games, midterms and Fed decisions, precisely the categories under legal attack in the United States. The venues fighting those battles hold federal licenses. Kalshi is a designated contract market; Polymarket operates domestically through an acquired exchange. Both submitted to registration, and both are being sued anyway.

World requires no brokerage account and no exchange registration, holds no customer funds, routes orders to liquidity providers on Solana, and settles in CASH, the dollar-backed stablecoin inside Phantom. That is a structurally different posture, and it raises a question the launch coverage has not touched: what happens when a non-custodial protocol lists the same contracts the licensed venues are being sued over. The optimistic reading is that a protocol holding no funds sits outside frameworks built around intermediaries. The skeptical reading is that offshore venues made the same argument before 2022, and it ended in a settlement and a geoblock.

The quiet concentration story

Follow the infrastructure and a different risk appears. World runs on Chainlink Data Streams and the Runtime Environment, and another prediction market adopted the same stack earlier this year, with plans to extend to stocks and commodities. Two venues on one settlement stack in a single year is how a standard forms, and a standard is how correlated failure arrives. A rulebook venue and an oracle venue fail independently. Two venues on the same data infrastructure do not.

None of this makes World’s design wrong. It makes it correct for a specific class of contract and unsuitable for another. The single most useful piece of due diligence on the platform is checkable from the market list: what fraction of those 150,000 markets involve contested definitions rather than clean data. If it is small, the design fits. If it is not, the first genuinely disputed settlement will arrive with no mechanism at all for handling it, which is worse than either alternative.