The full breakdown
What counts as a spread, and how the numbers are derived.
The homepage keeps this short by design. Here is the detail behind it.
The full breakdown
The homepage keeps this short by design. Here is the detail behind it.
What it detects
One event where exactly one listed outcome must win, a Polymarket winner-take-all event, Kalshi brackets that cover every possible result, or on Polymarket US a 3-way soccer result or a championship with every team listed. Buy YES on every outcome for under $1, or NO on every outcome for under the payout, and the payout is fixed, as long as every leg fills at those prices. Events where an unlisted result could win are skipped. On every plan, Base included.
Two brackets on the same number in one event, where one range contains the other: “4.00% or above” happens whenever “4.25% or above” does. Buy YES on the wider and NO on the narrower for under $1 and the dollar is fixed, whatever the number turns out to be. Two legs instead of five, and a single bracket carries far more resting size than a whole ladder. On every plan, Base included.
A Kalshi market and a Polymarket market that look like the same real-world event. Pairs are matched automatically when both platforms list the same question with the same wording, numbers and deadline, and priced with both platforms’ fees; you can hide any pair, or add one yourself. An automatic match can still hide a different rule, so check both sets of rules before you trade. Quant.
Why the list is short
Plenty of screens will show you two prices that don’t match. Most of those aren’t trades. Here is everything that has to hold first.
Exactly one listed outcome must win, or one bracket’s range must contain another’s. Events where an unlisted result could win are skipped even when the prices look free. The scanner sets aside thousands of events a sweep for exactly that reason.
Every leg is priced at the live best ask and the quantity actually offered there, then each platform’s own published fee formula comes off. A spread that clears before fees and not after never reaches the table. That is why the number you see is smaller than the ones screenshotted on social media.
One tap sends every Kalshi and Polymarket US leg as a fill-or-kill limit order, up to 40 legs, thinnest book first, signed in your browser with keys that never reach our servers. Nothing is sent until you confirm the ticket. A read-only screen stops at the row. Quant.
This is not the fastest scanner and doesn’t claim to be: the feed re-prices about once a minute, and dedicated bots react in milliseconds. What survives here is what they skip, on quieter venues, at sizes not worth their time, or with money tied up for weeks.
How this is different
Most arbitrage tools are screens: they compare prices, flag the difference and hand you a link. That is genuinely useful, and it stops three steps short of a position. Here is where the two part company.
It does one thing
ArbTrader only looks for arbitrage. There is no model here, no fair-value estimate, no rating, no pick, and no opinion about which way a market is heading. Every number on this site comes from two places: a structure that can be proved, and the live order book behind it.
That is deliberately a narrower product than a general betting tool, and the narrowness is the feature. A tool that ranks bets by expected value is only as good as the model behind the estimate, and you inherit whoever built it. Here there is nothing to inherit. You never have to trust our view of an event, because we do not have one: either exactly one listed outcome must win and the legs cost less than the payout after fees, or there is no row.
Two prices don’t match, so it shows you both.
Checks whether both prices can be wrong at once. It proves that exactly one listed outcome must win, or that one bracket’s range contains another’s, before the row exists. When it can’t prove it, the event is dropped.
Quotes the headline price at the top of the book.
Quotes the size actually resting at that price, on every leg, and takes each platform’s own published fee formula off before calling anything a spread. A gap that only exists before fees never appears.
Ends at a link. You retype the trade somewhere else, leg by leg, while the price moves.
Ends at a confirmed order. One tap sends every Kalshi and Polymarket US leg as a fill-or-kill limit order, up to 40 legs, thinnest book first, signed in your browser. Quant.
Where a screen wins: raw speed, and how many venues it watches. A dedicated terminal re-prices in about a second and follows a dozen venues; this re-prices in about a minute and follows three. If you want the widest possible view of the market, that is the better buy. If you want prediction-market spreads that are proven before you see them and placeable when you do, that is this.
What to expect · in detail
How big the edges are. A complete spread, where every leg fills, usually returns 1 to 3% after fees. Anything much bigger is usually a price that isn’t really there: a stale or frozen book, or a single stray order. ArbTrader now refuses those: a group’s prices have to add up to about $1 and every book has to be live.
What that means in dollars. At $10 to $30 a trade, a complete spread makes roughly 10¢ to 60¢. The complete spreads placed through ArbTrader in its first week each made between about 8¢ and 60¢ on $10 to $31 staked, and paid out whoever won. Fees are charged on every order and rounded up to the cent, so very small trades can lose money even when every leg fills; tickets refuse those.
How often they appear. It varies from day to day and hour to hour. Real spreads are rare and most close within seconds to minutes, often because other traders and bots take them first. Games that haven’t started yet move slowly and fill more reliably; live games move every second.
Where it can go wrong. Every leg is a separate order and no exchange offers “all legs or none”, so a price can move between orders and leave one leg unfilled. ArbTrader re-reads every book the moment you place, needs half again your size resting on each leg, retries a missed leg up to break-even, and otherwise takes whichever way out loses less: finishing the spread at a small locked loss, or selling back near the market price. A miss usually costs cents; it can cost more.
Why we took the projection down. This page used to project 90 days forward from one snapshot of the feed. Real trading showed that one person reaches far fewer of those spreads, at smaller sizes, than a snapshot suggests, so the projection overstated what to expect. We’d rather show you real numbers than a flattering line.
For comparison. Most people who bet on outcomes on these venues lose money: a Wall Street Journal analysis of Kalshi’s data in 2026 found about 74% of accounts finish at a loss. A checked spread doesn’t depend on who wins, which is the whole point. Nothing here is a guarantee, a past result you can count on, or advice.