How to Split Stakes on Both Sides of a Prediction Market Arb (Polymarket vs Kalshi)
The equal-contract math that locks a payout across Polymarket and Kalshi, worked fee-by-fee, plus the traps that unmake a lock: thin books and mismatched resolution criteria.
Why do Polymarket and Kalshi price the same event differently?
Polymarket and Kalshi are two venues pricing many of the same real-world events, and they disagree constantly. One is an on-chain market settled in crypto rails, the other a regulated US exchange, so they draw different crowds, run different fee structures, and impose different frictions on moving money in and out. Capital does not flow freely between them, no one can cross-margin a position across both, and plenty of traders can only access one side. The result is that the same outcome routinely trades cents apart on the two venues for hours or days at a time.
Because every share on either venue pays exactly $1 if its outcome happens and $0 if it does not, a price gap between venues is not an aesthetic disagreement, it is a measurable object. If YES on one venue plus NO on the other can be bought for less than $1 combined, all-in after fees, the difference is capturable whichever way the event resolves, subject to real caveats this guide spends its second half on. That structure, buying both sides of the same question on different venues below the guaranteed payout, is the cross-exchange arb.
The Divergence & Arbitrage scanner at /divergence hunts for exactly these pairs, matching equivalent markets across the two venues and surfacing the gaps wide enough to matter, and its built-in stake splitter does the sizing arithmetic this guide walks through. The scanner deliberately only pairs events it is confident are the same market, because a false match is worse than a missed one, a point that becomes very concrete in the resolution-criteria section below. Nothing here is financial advice, and only trade money you can afford to lose.
How does the equal-contract split lock a payout?
The whole trade rests on one principle: equal contracts, not equal dollars. Buy N YES contracts on one venue and N NO contracts on the other, and exactly one of those two positions pays out, whichever way the event resolves. N shares of the winning side at $1 each means the combined position returns exactly $N in every state of the world. The payout is fixed by the contract count alone, which is what makes the position a lock rather than a directional trade.
The intuitive mistake is splitting a budget into equal dollar halves. Because the two sides trade at different prices, equal dollars buy unequal contract counts, and unequal counts leave you with a residual directional position dressed up as an arb: fine when the event goes one way, underwater when it goes the other. The splitter sizes by contracts, computing the all-in cost of one contract on each side and dividing your budget by the sum, flooring to whole contracts. It also tries both directions, YES on Polymarket with NO on Kalshi and the reverse, and takes whichever combination is cheaper all-in, since fees can make one routing viable and the other not.
The viability test is a single number: the cost per $1 of guaranteed payout, which is YES price plus YES fee plus NO price plus NO fee. Below $1 and the lock clears; at or above $1 there is no trade, no matter how wide the raw price gap looked before fees. The splitter reads its budget from the same shared bankroll settings as the Kelly calculator on /bankroll, so your arb allocation and your directional sizing come from one coherent set of risk numbers rather than two separate guesses.
What does a real split look like, fee by fee?
Take a realistic pair: YES trades at 40¢ on Polymarket while NO trades at 55¢ on Kalshi, which is to say Kalshi prices YES at 45¢, a five-cent disagreement on the same event. The Polymarket leg costs 40¢ plus roughly 0.2¢ of gas and slippage, about 40.2¢ all-in. The Kalshi leg costs 55¢ plus the exchange fee of 0.07 × 0.55 × 0.45, about 1.7¢, so roughly 56.7¢ all-in. One contract of each side therefore costs about 96.9¢ and pays exactly $1 whichever way the market resolves.
With a $200 budget the splitter buys 206 contracts on each side, $200 divided by 96.9¢, floored to whole contracts. That puts about $82.81 on the Polymarket YES leg and about $116.87 on the Kalshi NO leg, roughly $199.68 in total, for a guaranteed $206 payout: about $6.32 of locked profit, a 3.2 percent return on the capital committed, earned identically in both outcomes. Note how lopsided the legs are, the NO side carries 41 percent more capital than the YES side, which is exactly the equal-contract principle at work; equal dollars here would have left a directional stub.
It is worth staring at what the fees consumed. The raw gap was five cents per contract; the two legs' fees ate about 1.9¢ of it, leaving roughly 3.1¢ of net edge. Nearly two-fifths of the apparent arb went to costs before any execution risk entered the picture, and a two-cent gap in the same market would have died entirely. This is the general shape of prediction-market arbitrage: small, fee-bitten edges that only exist after honest accounting, which is precisely why the splitter prices every fee into the viability check rather than letting the sticker gap flatter the trade.
What eats the edge: fees per leg and thin books?
Fees ride on every leg, and an arb by definition has two of them. Kalshi's 0.07 × P × (1 − P) fee is heaviest exactly where arbs are most common, near the middle of the price range, peaking around 1.75¢ per contract at 50¢. Polymarket's roughly 0.2¢ per share is smaller but never zero. A gap has to clear the combined fee load of both legs to be worth anything, and gaps of a cent or two, which are the ones you will see most often, usually do not.
The quoted price is the top of the book, and the book has depth. Buying 206 contracts into a market whose visible liquidity is thinner than your order walks the price against you as you fill, and a lock computed at quoted prices can compress toward zero by the time both legs are done. The scanner flags splits that exceed the Polymarket book's depth so you can size down before entering rather than discover the slippage afterward. In thin markets the honest arb is often a fraction of the budget the raw math would allow.
Two frictions remain even with fees and depth respected. Legging risk: the two orders do not fill simultaneously, and the price that made the pair attractive can vanish between leg one and leg two, leaving you holding a single directional position you never wanted. And capital duration: both legs stay funded until the market resolves, so a 3 percent locked return is a very different proposition on a market that settles this week than one that settles next year. An arb is a real trade with real execution, not a coupon you clip.
What is the resolution-criteria trap?
The most dangerous assumption in cross-venue arbitrage is that two markets with the same title are the same market. The famous burn is a government-shutdown market that resolved YES on Polymarket and NO on Kalshi: same headline event, but the two venues had written different resolution criteria, different definitions and thresholds for what officially counted, and the same night graded in opposite directions. Traders who held what they believed was a riskless pair discovered that the lock only ever existed inside a single rulebook.
Run the failure case through the math and it stops being abstract. The equal-contract lock pays $N because exactly one of your two positions wins. If the venues resolve in opposite directions, that guarantee is gone: hold the fortunate pair of sides and both legs pay, but hold the other pair and both legs lose, turning a computed 3 percent lock into a loss of roughly your entire committed capital. There is no hedge after the fact, because the divergence happens at settlement, when both positions are beyond adjusting.
The discipline is unglamorous: read both venues' full rule text before splitting a single dollar, and treat any mismatch in sources, deadlines, definitions, or edge-case handling as a no-trade regardless of how wide the gap is. Ambiguous physical-world events, shutdowns, ceasefires, resignations, anything graded off interpretation rather than a single published number, deserve the most suspicion, and a gap that looks too generous is often the market's way of pricing exactly this risk. The splitter locks the arithmetic; only you can verify the two rulebooks describe the same world.
Can you arb inside a single venue? Overround and the companion scanner
The single-venue variant of the same trade is intra-market overround. In a multi-outcome event whose outcomes are mutually exclusive and exhaustive, the complete set of YES shares must pay exactly $1 to whichever outcome occurs, so if the whole set can be bought for less than $1 after fees, buying every outcome locks the difference, no second venue required. The mirrored construction on the NO side works the same way against its own fixed payout. These gaps appear most often in fast-moving multi-candidate markets, where prices update outcome by outcome and briefly stop summing to a coherent book.
The structural advantage of the single-venue lock is that the resolution-criteria trap disappears: one venue, one rulebook, one grader, so the settlement risk that haunts cross-venue pairs simply is not there. The costs move elsewhere. A five-outcome overround is five legs, each paying its own fee and crossing its own spread, and the thinnest outcome in the set governs how many complete sets you can actually assemble at the quoted prices. The arithmetic discipline is identical to the two-venue case: total all-in cost per $1 of guaranteed payout, and no trade at or above the line.
The Overround Scanner at /overround is the companion tool that watches for those incoherent sets inside each venue, while /divergence covers the cross-venue pairs, and both draw on the same shared bankroll settings so your sizing stays consistent across every lock you consider. Kept in proportion, arbitrage in prediction markets is a scavenger's craft: small edges, honest fees, careful execution, and rulebooks read in full. None of this is financial advice or a promise of profit, past performance does not guarantee future results, and even a lock deserves only capital you can afford to lose.
WhaleTracks is informational analytics, not financial advice. Past performance does not guarantee future results.