04 · The Oracle · Kalshi · Polymarket
Kalshi and Polymarket price the same futures on different rules. The gaps are real: two to four cents on the dollar between matched questions, and three to seven cents for weeks on the 2024 election. The hard part was never finding the gap. It is proving the two questions are the same question. The Oracle reads both books, prices every gap net of fees, and tells you which ones survive the guards.
01 · Two books, one question
Buy one side on Kalshi and the other on Polymarket. If the two legs cost less than a dollar after fees, the pair pays no matter what happens, provided both markets resolve the same way. Strict means the wording, strikes, dates and shape all line up. Subset means one question is a stricter version of the other, so YES on the looser one plus NO on the stricter one covers every outcome. Review means something does not line up, and the note says what.
| Label | Edge | Kalshi leg | Polymarket leg | Cost | Match |
|---|
Before any capital: read both rulebooks. Kalshi settles by its Outcome Review Committee. Polymarket settles by UMA's optimistic oracle. The same words can end differently.
02 · One book, both sides
YES plus NO on the same market, same book, same rules. No resolution risk at all. A well-made book sits at $1.02 to $1.06 for the pair, so a true lock is rare and short-lived. These are the closest to it right now.
| Kalshi | Yes | No | Fees | Edge |
|---|
| Polymarket | Yes | No | Fees | Edge |
|---|
03 · One of these must happen
On a Kalshi event where exactly one outcome wins, buying YES on every leg pays a dollar. If the legs cost less than a dollar after fees, the whole list is a lock. Only events the book itself treats as complete are shown, meaning the YES bids across the legs sum near a dollar.
| Event | Legs | Sum of asks | Sum of bids | Fees | Edge |
|---|
04 · How the Oracle reads
Prices across prediction markets do not converge because the contracts are not fungible: the same event can resolve differently on two venues, so nobody can net a position across them. That is a structural gap, not a market failure. Gross price gaps are common. Provable equivalence is the scarce thing.
About 6% of events are listed on more than one venue, but they are the long-lived ones and carry a tenth of all event-days. Roughly 8% of Polymarket's markets have a twin elsewhere against 2% of Kalshi's. Matched twins sit 2 to 4 cents apart on average, and the gap holds for weeks.
Gebele & Matthes, arXiv 2601.01706. Ten platforms, 102,275 events, 2018 to 2025.
Both venues charge more when the price is near fifty cents and almost nothing at the tails.
Kalshi: 0.07 × p × (1 − p) per contract, taker.
Polymarket: rate × p × (1 − p) when the market has fees on, zero when it does not. The rate is read per market.
Every edge on this page is net of both.
Two questions are matched by weighted word overlap, then checked: strikes must agree, dates must agree, the shape must agree (win, finish third, above a number, between two numbers), rate words must not oppose (cut versus hike), and neither side may carry a qualifier the other lacks (vice, senate, primary, runoff). Fail one and the pair is marked review. Weather is always review: Kalshi settles New York on Central Park and Polymarket on LaGuardia, and the paper's own pipeline got that one wrong. Sports were outside the paper's study, so treat those rows as unproven.
Kalshi trade API v2, open events with nested markets. Polymarket Gamma API, active binary markets ordered by 24-hour volume. Both public. The Oracle only reads. It places nothing, holds nothing, and links you to the two markets so you can read the rules yourself.