Methodology
Why arb scanner numbers are wrong in 2026
Kalshi-Polymarket arb scanners disagree because they measure different things: raw mids, executable books, round-trip taker fees, close times, and rule forks. Kalshi charges takers 0.07 x P x (1 - P); Polymarket charges category rates from 0.00 to 0.07. TapeXray’s board ranks every match at that 0.07 maximum, then re-prices gated pairs at their real rate.
What this page found
- The live board rejected 100 suspect pairs before ranking gaps, because bad matches create cleaner-looking arbs than real markets do.
- Only 2 gated pairs were positive after fees, while 4 pairs sat inside the fee line on the 2026-09-21 17:30 UTC board.
- Kalshi’s published taker fee is 0.07 x P x (1 - P) per contract per side, with maker fee listed as 0.0.
- Polymarket’s published taker rates run from 0.0 on geopolitics to 0.07 on crypto, with maker fee listed as 0.0.
- TapeXray’s live gap math ranks every match at a conservative 0.07 Polymarket maximum, then re-prices gated pairs at their real category rate, so the board is a screen, not a final execution model.
Why do arb scanners disagree?
Most Kalshi-Polymarket scanner disagreements start with the same mistake: treating a displayed price gap as an executable edge. A raw 81.0c versus 73.5c midpoint gap is not the same thing as a filled trade. The trader does not get the midpoint by right. The trader crosses the spread, pays taker fees if lifting offers or hitting bids, and then carries rule, timing, and resolution risk.
TapeXray’s live board is built to be quotable, not flattering. It gates candidate matches, rejects suspect pairs, subtracts a fee line, and separates ordinary midpoint gaps from locked forks that can be priced off the book. The result is usually smaller than the number a raw scanner prints. Some days the answer is that most days there is nothing here.
What does TapeXray subtract before calling a real gap?
The midpoint gap is only the first cut. TapeXray’s board uses fee_adjusted_cents = |mid_k - mid_p| - 200 x (0.07 x Pk x (1 - Pk) + rp x Pp x (1 - Pp)). Pk and Pp are the Kalshi and Polymarket mids in dollars, and rp is the Polymarket rate - 0.07 while ranking and gating the whole match set, then each gated pair’s own published category rate once the gate has narrowed roughly 10,900 matches to a few dozen. The 200 multiplier is there because that screen treats the comparison as a round trip, entry plus exit, on both venues.
That two-pass design is intentionally conservative for a scanner, but it is not the same as a fill ticket. A locked fork is priced differently. For fork_net_cents, the board prices both legs off the book: YES ask on the cheap venue and NO = 100 - YES bid on the other venue. It then subtracts one-way taker fees per leg, because a fork held to resolution has no exit leg.
| Item | Board treatment | Why it matters |
|---|---|---|
| Midpoint difference | |mid_k - mid_p| | Shows the visible gap before costs |
| Round-trip fee line | 200 x (0.07 x Pk x (1 - Pk) + rp x Pp x (1 - Pp)); rp is 0.07 ranking, the real rate once gated | Removes estimated taker costs on entry and exit |
| Fork pricing | YES ask plus NO from 100 - YES bid | Uses executable book sides instead of mids |
| Fork fees | One-way taker fees per leg | A held fork settles; it does not need an exit leg |
How do Kalshi and Polymarket fee schedules differ?
Kalshi and Polymarket do not publish the same taker-fee schedule. Kalshi’s verified formula is fee = ceil(0.07 x C x P x (1 - P)) cents, where C is contracts and P is the price in dollars. The maker fee is listed as 0.0. The fee is charged to takers on both entry and exit.
Polymarket’s verified formula is fee = C x rate x P x (1 - P), with the rate set by market category. Its maker fee is also listed as 0.0. That category step is why a single flat Polymarket estimate can be useful for a screen but wrong for a specific bet.
The P x (1 - P) shape matters. Kalshi’s note says the fee peaks at 50c, where 0.07 x 0.5 x 0.5 = 1.75c per contract per side, and falls toward zero at either extreme. Polymarket uses the same shape, but the category rate changes the cents paid.
| Venue | Published taker fee | Maker fee | Verified |
|---|---|---|---|
| Kalshi | ceil(0.07 x C x P x (1 - P)) cents | 0.0 | 2026-09-22 |
| Polymarket | C x category rate x P x (1 - P) | 0.0 | 2026-09-22 |
| Category | Rate |
|---|---|
| crypto | 0.07 |
| culture | 0.05 |
| economics | 0.05 |
| finance | 0.04 |
| geopolitics | 0.0 |
| mentions | 0.04 |
| other | 0.05 |
| politics | 0.04 |
| sports | 0.05 |
| tech | 0.04 |
| weather | 0.05 |
What still makes the TapeXray number wrong?
The TapeXray number is still an estimate. It can be a better estimate than a raw midpoint screen and still be wrong for a trader’s actual fills. The board cannot guarantee the displayed size is available when the order arrives. It cannot guarantee the other venue still shows the matching leg. It cannot make two rulebooks identical.
- Stale quotes: a price can update on one venue before the other, making a short-lived gap look cleaner than it is.
- Spread crossing: midpoint math does not show the bid and ask a trader must actually hit.
- Close-time mismatches: two markets can describe the same event but stop trading on different clocks.
- Rule forks: wording can diverge enough that both legs are not true economic opposites.
- Fee schedule mismatch: only a gated pair is re-priced at its real Polymarket category rate. A rejected suspect pair is never checked against its true rate and keeps the conservative 7% maximum.
This is why the Kalshi vs Polymarket fee guide belongs next to the board, not behind it. A board number can identify a candidate. It does not replace line shopping, reading the contract terms, checking the last trade, and modeling the exact taker fee for the category.
What did the latest board actually show?
The 2026-09-21 17:30 UTC board tracked 236,027 markets across two venues. It listed 94,209 active Kalshi markets and 138,833 active Polymarket markets in the catalog. That large catalog does not mean there are many clean cross-venue bets. After the gate, the board showed 6 gated pairs, 4 inside the fee line, and 2 positive after fees.
The top displayed gap was a politics market: “Will Elif Eralp win the election for Governing Mayor of Berlin by the Berlin House of Representatives following their state election?” The Kalshi midpoint was 81.0c, the Polymarket midpoint was 73.5c, and the fee-adjusted gap was 3.8c once that market’s own 0.04 politics rate replaced the 0.07 ranking maximum. The second displayed gap was an other-category chess market at 60.0c on Kalshi, 51.5c on Polymarket, and 2.6c after the board’s fee adjustment at its 0.05 sports rate.
Those are board numbers, not a promise that a trader can lock the same return. The next check is the market page, the order book, the fee category, and the rules. TapeXray’s market archive preserves the tape; the methodology notes explain how to cite it without overstating it.
| Market | Category | Kalshi mid | Polymarket mid | Polymarket rate | Fee-adjusted gap |
|---|---|---|---|---|---|
| Will Elif Eralp win the election for Governing Mayor of Berlin by the Berlin House of Representatives following their state election? | Politics | 81.0c | 73.5c | 0.04 | 3.8c |
| Will Uzbekistan win the 46th FIDE Chess Olympiad Open Tournament? | Other | 60.0c | 51.5c | 0.05 | 2.6c |
Questions people ask
Why does one scanner show a bigger Kalshi-Polymarket arb than another?
Usually because the scanners are not measuring the same thing. One may use raw midpoints. Another may cross the spread. TapeXray subtracts a round-trip fee line using 0.07 for Kalshi and, for each gated pair, that market’s own Polymarket category rate, then rejects suspect pairs before ranking gaps.
Is TapeXray’s gap number executable?
No. It is a board estimate, not a fill guarantee. It can be useful for finding candidates, but the actual trade depends on live bids and asks, available size, category fees, timing, and whether the Kalshi and Polymarket rulebooks are true opposites.
Why does TapeXray rank matches at a 7% Polymarket maximum first?
Reading every one of roughly 10,900 matched pairs’ published category rate up front would overrun the board’s memory budget, so the whole match set is ranked and gated at 7%, the highest published rate. Only the pairs that pass the same-bet gate are then re-priced at their own category rate, 0.0 to 0.07, so a gated pair’s number can only rise on re-pricing, never fall.
Do fees really erase most displayed gaps?
Often, yes. On the 2026-09-21 17:30 UTC board, 4 pairs were inside the fee line and 2 were positive after fees. A visible midpoint difference is common. A clean, fee-adjusted, executable cross-venue position is less common.
Where should a trader check next after seeing a gap?
Check the live order books, contract rules, close times, fee category, and available size before treating any board gap as tradable. The board is the first filter. It is not a substitute for line shopping or reading both venue rulebooks.
Sources
Independent archive. No referral links in editorial copy, no paid placements, and nothing here is legal, tax or financial advice. Found an error? Tell us and we will date the correction.
Related reading
See today's board
Every pair on this page is scored live against the same fee line, refreshed with each deploy. No referral links, no rankings we were paid for.
Free channel: gaps at the next rebuild. Members: the instant one matching your watches, in your DM.
Get alerts (opens in a new tab)