Two weather stations
“Will New York hit 90°F?” One app reads the Central Park station, the other LaGuardia airport. On the same afternoon they can differ by a few degrees, so both of your answers can lose.
The Oracle · How we check
Sometimes two apps sell the same question at different prices. Buy both answers and one of them pays $1. Most of the time, after fees, there is nothing left. Here is every check a pick goes through before you see it, in plain words.
Kalshi, Polymarket and Polymarket US often list the same question: who wins a game, who wins a vote, who wins an award.
The Oracle compares three apps: Kalshi, Polymarket and Polymarket US, a separate app with its own prices, fees and rules. A pick can use any two of the three apps, so you need an account on both apps it names. Each app sets its own rules on who can trade there.
We compare every open question on one app with every open question on the others and keep the ones that look alike. Then we read both apps’ rules side by side: what counts, who decides the result, and on which day. A pair only reaches the page when the rules say the same thing, or when they differ in one rare case that we can name for you.
By our estimate from a sample, about 1 in 5 Polymarket questions that end in the next two months has the same question on Kalshi. We are testing a computer program that reads both apps’ questions and suggests more matches. Our strict rule check always has the final say: in our test it caught 3 of 49 “same question” calls that were wrong on the rules.
Two questions can read the same and still pay differently.
Almost always the same answer. Not always.
A written post can make one answer win on one app and lose on the other.
When the rules differ in one rare case, the card names it under Could go wrong, with what you would get back in that case at your amount and a rough chance, marked “our estimate” when it is our judgment. When we can’t name the case, we don’t show the pick.
Both apps take a small fee when you buy. It is biggest on coin-flip questions, priced near 50¢, and small near 1¢ or 99¢.
On a coin-flip game the two fees add up to about 3¢ per $1. That is about the size of a typical gap, so most gaps vanish once you pay them:
Before fees this pair costs 97¢ and pays $1: it looks like 3¢ of free money. After both fees it costs about $1.00, and nothing is left. Every number on the Oracle is after both fees.
On 25 Sep, in one check of 1,124 games listed on both apps, 13 still showed a gap after fees. 9 of those games had already started, 1 was a mismatch and 1 was gone on a second look, so only 2 were real before the game began. The Oracle page shows the count from its own latest check.
Each app sells only so much at the best price. After that the price goes up and the gap shrinks.
We count how many pairs you can buy before the gap is gone. That is the “Only $… is for sale at this price” line on a card. Putting in more than that does not earn more: the Oracle works out your pick at the amount that fits and tells you so.
We also check the smallest order each app takes, so a pick never asks you for less than an app will sell.
Fast gaps often last only minutes. A typical short one we saw lasted about 11 minutes, and half were gone before the next half-hour check.
We check all three apps about every 30 minutes. In between, we re-check the price of every pick we can watch live (almost all of them), and of every game listed on two apps that starts in the next 12 hours, every few seconds. A new gap we catch in between shows up on the Oracle within seconds, marked Just found. While you have the Oracle open, the prices on screen refresh every few seconds. Each card says when it was last checked:
Checked 3 s ago · green: fresh, you can act.
Checked 4 min ago · grey: probably still there.
Checked 12 min ago. Too old to trust · orange: the buy buttons are off until a new check.
On picks that pay within a week, the buy buttons only work when the price is at most 15 seconds old and has passed two checks at least 5 seconds apart. Games already being played are never shown: every gap during a game that we re-checked was gone a moment later, 17 of 17.
These pairs look like a gap on paper. We skip them, or take them only when both apps’ rules match.
“Will New York hit 90°F?” One app reads the Central Park station, the other LaGuardia airport. On the same afternoon they can differ by a few degrees, so both of your answers can lose.
On the 15-minute Bitcoin questions, one app pays on an average of the last prices, the other on the single closing price. By our calculation they disagree about 1 time in 6.
The apps take the price at different times of day and from different price sources. The “same” question is really two questions.
“Will they say the word?” When one app counts written posts and the other only words said out loud, we skip it. We only take these when both apps count the same thing, such as words said out loud at one named speech at the UN. That is where our first finished picks came from.
Prices jump every few seconds once a game starts. Every gap during a game that we re-checked had turned into a loss.
A pair that pays in 2028 can earn less than savings: about 1.6% a year, while a US Treasury bill pays 4.04%. The Oracle folds these under Long waits and says so on the card.
Advanced
Seen two prices yourself? Type them in and see what the pair pays either way after both fees, how many to buy on each app, and when it pays.
Check any two prices (advanced)Where the idea comes from. A 2026 study by Gebele and Matthes compared Kalshi and Polymarket and found the same questions often sat 2 to 4 cents apart for weeks. Its big returns came from 2022 to 2024, when the apps charged little or no fees. Today both apps charge, which is why most gaps don’t survive our fee check. Read the study.