No. We pulled 200 straight 15-minute Bitcoin windows on Kalshi, and the 66 that finished $50 or more from their starting price never flipped in the last minute. Not one. All 31 late flips came from windows that finished within $50 of the line, and 28 of those finished within $25. If you held a 91-cent Yes with two minutes left and watched it settle No, nobody took it from you. You were holding a coin flip that was priced like a lock. I have rerun the test twelve times since, on 200 fresh windows each time, most recently on October 26: across all thirteen samples, 13 wide flips in 1,423 wide windows, and none in the newest 200, which ran from early Saturday morning to early Monday morning. In all thirteen flips the move that made the window wide came inside the final two minutes. What Moved, below, walks through the window from this weekend that came closest to the complaint, a Yes at 90.7 cents on Sunday afternoon that settled No by $5.87 after Bitcoin gave back a $16 cushion in the minute after the two-minute mark.
| Price | What it says | |
|---|---|---|
| Wide windows | 0 flips | 66 windows finished $50 or more from the line. None flipped late. |
| All late flips | 31 | Every one came from a window that finished inside $50. 28 finished inside $25. |
| Typical margin | $33 | The median gap between the settled average and the line, on an $80,000 asset. |
| Settlement | 60 prices | The last minute's index prices, averaged. One wick is one sample of 60. |
The 200 windows that settled between 3:00 p.m. ET on September 4 and 4:45 p.m. ET on September 6, 2026: 98 Yes, 102 No, every one the "BTC price up in next 15 mins?" contract. Anyone can rerun it from the markets endpoint of Kalshi's public API. Rerun on September 21, September 26, September 29, October 2, October 5, October 8, October 11, October 14, October 17, October 20, October 23 and October 26, 2026, on the newest 200 settled windows each time. Those numbers are in What Moved, below.
Wide margins almost never flip: 13 in 1,423. Coin flips do. 2,600 windows in thirteen samples, September 4 to October 26, 2026. Thirteen wide flips, nine of them between October 6 and October 23, and each one was Bitcoin moving.
What Moved Since October 23
The numbers above come from 200 windows on September 4 to 6. I reran the test on September 21, September 26, September 29, October 2, October 5, October 8, October 11, October 14, October 17, October 20 and October 23, and I ran it again on October 26, on the 200 windows that settled between 1:45 a.m. ET on Saturday, October 24 and 3:30 a.m. ET on Monday, October 26. It was a quiet weekend for Bitcoin. Every settlement in the sample landed between $79,294 and $80,858, the typical finishing margin fell to $52 from $69, and 105 windows finished $50 or more from their line, 63 of them between $50 and $100 and 42 past $100. Every one of the 105 settled on the side the market was on with two minutes left. Nothing in it changed the answer. The 20 flips all came from windows that finished inside $50, 18 of them inside $25, and the closest the weekend came to the complaint was a Yes at 90.7 cents on Sunday afternoon that settled No by $5.87.
| Measure | Sept. 4-6 | Oct. 18-20 | Oct. 21-23 | Oct. 24-26 |
|---|---|---|---|---|
| Windows that finished $50 or more from the line | 66 | 107 | 123 | 105 |
| Late flips in those wide windows | 0 | 0 | 1 | 0 |
| Late flips, all 200 windows | 31 | 18 | 15 | 20 |
| Flips from windows that finished inside $25 | 28 | 16 | 10 | 18 |
| Flip rate among windows that finished inside $50 | 31 of 134, 23% | 18 of 93, 19% | 14 of 77, 18% | 20 of 95, 21% |
| Flips where the losing side was 70 cents or better with 2:00 left | 23 of 31 | 11 of 18 | 10 of 15 | 11 of 20 |
| Median finishing margin | $33 | $55 | $69 | $52 |
| Median contracts traded per window | 1.9 million | 2.9 million | 3.2 million | 2.9 million |
Same method in every column. A flip is a window whose result went against the side Yes was trading on with two minutes left, and the margin is Kalshi's published settlement value minus the window's line. Each rerun is the newest 200 settled "BTC price up in next 15 mins?" windows pulled from Kalshi's public API, the same contract the first sample used. Nine earlier reruns came out of the table to keep it readable on a phone: September 19 to 21, September 24 to 26, September 27 to 29, September 30 to October 2, October 3 to 5, October 6 to 8, October 9 to 11, October 12 to 14 and October 14 to 17. Together they were 1,022 wide windows with one wide flip in each of the first four, none in the fifth, one in the sixth, two in the seventh, four in the eighth and one in the ninth, 179 flips in all, 167 of them from windows that finished inside $50 and 137 from inside $25, and median margins of $55, $57, $83, $79, $36, $70, $34, $78 and $75. The October 21 to 23 column is 104 in a row, then a gap, then 96 in a row, because Kalshi's public API lists no windows closing between 3:15 and 5:00 a.m. ET on Thursday, October 22. The October 14 to 17 rerun, now in this paragraph, had the same Thursday-morning hole on October 15. The other three columns have no gap, and no window appears in two samples.
That Sunday window closed at 2:45 p.m. ET on October 25, on a line of $80,574.55. Yes spent the window's first five minutes between 46 and 54 cents, with Bitcoin between $2 and $11 under the line on Coinbase, then climbed to 70 cents by 2:40 as Bitcoin crossed over and reached $29 above the line. Yes printed 89 cents at 2:41, and with two minutes left it stood at 90.7 cents, with Bitcoin at $80,591.18, $16 over the line. The 2:43 minute, the one that began at the two-minute mark, opened there and closed at $80,544.54, $30 under, a $46 slide against a $16 cushion. The settlement minute opened $5.96 under, never got above the line, and closed $6.82 under, and Kalshi's published 60-price average landed at $80,568.68, No by $5.87. The contract tracked it fill by fill: Yes traded at 93 cents with 100 seconds left, 45 cents with 90 seconds left, 10 cents with a minute left, 3 cents with 30 seconds left, and a tenth of a cent through the final 10 seconds. The last fill over 50 cents printed with about 84 seconds left, and the contract never traded above 50 cents again. About 1.36 million of the window's 3.32 million contracts traded after the two-minute mark. A 90.7-cent Yes on a $16 cushion is the complaint in one window, and it sits a hair under the 91-cent bar the top of this page uses, which is the point about bars: the 64 windows that had Yes at 91 cents or better with two minutes left all won, the one that had Yes at 90.7 cents lost, and nothing separates them except where I drew the line.
That keeps the count at 13 wide flips in 1,423 wide windows, about 1%, against 250 in 1,177, about 21%, for windows that finished inside $50. Nine of the thirteen came between October 6 and October 23, nine of 794 wide windows in the last seven samples against four of 629 in the first six, and this weekend shows the other half of that sentence. Bitcoin moved less, so fewer windows finished wide, 105 against 123 the sample before. The biggest margin, a Saturday 11:15 a.m. window that settled $457 under its line, was decided early: a $217 minute put Bitcoin more than $570 under the line ten minutes into the window, Yes was already at a penny, and it was a fifth of a cent with two minutes left. The October 21 flip, a 97-cent No that Bitcoin ran through with a $170 minute, and the $306 window on October 16, still the biggest of the thirteen, both needed that kind of minute late, against a confident price, and the $217 minute this weekend produced came ten minutes in, with nothing left to decide. Wide is a label you can apply once the window is over, not a filter you can trade at the two-minute mark. A window that is already wide with two minutes left is usually over, and when the price is confident on a thin cushion, as it was at 2:45 on Sunday, the price is no protection.
The shape holds. All 20 flips finished inside $50, 18 inside $25, and in 11 of the 20 the losing side was 70 cents or better with two minutes left. Sixty-eight windows had No at 91 cents or better with two minutes left, and two lost, both on Bitcoin running back to a line it had been well under: a Saturday 7:00 p.m. window with Yes at 5.3 cents that settled Yes by $3.13, and the midnight window going into Monday, which had Yes at 8.8 cents and Bitcoin $53 under the line at the two-minute mark. In the minute that followed, Bitcoin closed $17 over and Yes went to 67 cents; the settlement minute then slipped back to $10 under by its close, and the 60-price average still landed $5.15 over, on a window whose 4.84 million contracts were the third-most of the 200. That is the averaging rule cutting the other way, a price that finished against the holder and won anyway. With the October 21 to 23 sample, that makes four No-side misses at 91 cents or better in two samples against none on the Yes side, which looks like a pattern until you move the bar and count. At 85 cents or better across the same 400 windows, 150 Yes windows lost four times and 152 No windows lost six times, and in this weekend's 200 it was three and three. Two samples is too few to call that anything. The closest finish came at 8:45 p.m. on Saturday: a line of $80,765.43, a settlement of $80,764.48, No by 95 cents of Bitcoin, with Yes at 69 cents two minutes out on a cushion of $3.90, and 5 cents a minute later. And the heaviest final minute of the 200, 1.67 million contracts, came on a Sunday 9:15 p.m. window that was never in doubt, Yes at 2.5 cents with two minutes left and $64.68 under at the close, while the other five of the six heaviest final minutes all came on windows that finished inside $30. The programs pile into windows sitting on the line, and this weekend the single heaviest minute went to one that was already over.
The Proof
Five windows from the sample, at the marks a trader actually watches. Every one finished within $5 of its line, and the last column is the only one that decided anything.
| Window (ET close, Sept. 4-6) | Yes At 2:00 Left | At 1:00 | At 0:30 | At 0:10 | Settled | Average Vs Reference |
|---|---|---|---|---|---|---|
| Fri 5:45p | 61¢ | 39¢ | 29¢ | 42¢ | Yes | +$0.63 |
| Sat 8:00a | 91¢ | 25¢ | 8¢ | 2¢ | No | -$4.14 |
| Sat 2:45p | 10¢ | 41¢ | 95¢ | 89¢ | Yes | +$2.41 |
| Sun 8:15a | 2¢ | 1¢ | 45¢ | 30¢ | No | -$0.66 |
| Sun 12:15p | 83¢ | 18¢ | 57¢ | 0.6¢ | No | -$4.50 |
Last trade price for Yes at each mark, from Kalshi's public trade history. "Average Vs Reference" is the settled 60-second average minus the window's opening price, in dollars of Bitcoin.
The Sunday 8:15 a.m. window is the one I would screenshot if I wanted to prove a conspiracy. Yes traded around a penny going into the final 50 seconds, ran to 60 cents, faded to 23, spiked to 93 cents inside the final 20, and printed its last trade at a tenth of a cent, with roughly 725,000 contracts changing hands in those 50 seconds. It settled No, 66 cents of Bitcoin price under the line. That is not a result being taken away. That is a crowd repricing a running average that was sitting on the line, in public, at speed.
Two more numbers from the 200. Of the 74 windows that finished within $25 of the line, only 11 were trading between 40 and 60 cents with two minutes left, and 31 were already past 85 cents or under 15. At two minutes the price barely tells a coin flip from a decided window. And 23 of the 31 flips were priced at 70 cents or better for the side that ended up losing. Seventy cents was never a safe lead.
Why It Felt Rigged
What decides itthe average of 60 index prices in the final minute, published by CF Benchmarks, an outside index provider that builds its number from crypto exchange data. A whale hitting Kalshi's book in the last seconds changes who paid what for contracts. It does not move the number the market settles on.
Kalshi's own rules for the series say it plainly: "At the last minute before expiration, 60 RTI prices are collected. The official and final value is the average of these prices, rounded to the nearest 2 decimal places." The window's opening price is built the same way, from the 60 seconds before the open, and a tie goes to Yes. The full terms are in Kalshi's CRYPTO15M contract document.
That averaging rule is why the last minute feels like a robbery. The final 15 seconds are only a quarter of the average, so closing a gap in the running average takes a move about four times the size of the gap. You can watch Bitcoin rip $200 through the line in the last stretch and still lose, because 45 of the 60 samples were already locked in on the other side. It cuts the other way too. Sometimes the price finishes against you and you win anyway, because the average is the product and the last print is trivia.
The second reason is speed. The Sunday 8:15 window printed 864, 1,185, 1,281, 1,248 and 911 fills in its five final 10-second buckets. A decided window the night before, settled $198 above its line, printed 92 fills in its entire final minute. Same product, same minute on the clock, about 60 times the activity when the average sits on the line. Nobody clicks that fast. Programs reading the public index reprice the contract before you can, and what you experience as a flip is them getting there first. That is a speed advantage, and it feels unfair. It is not a corrupted settlement.
The third reason is arithmetic. These windows resolve four times an hour, 96 times a day. Near the line, a 70-cent contract is honestly a 70-cent contract, roughly a -233 favorite, and -233 favorites lose all the time. Hundreds of "I was winning and it flipped" stories get minted every week, and the losers post about it while the winners quietly collect a dollar a contract.
What Would Change My Mind
I cannot certify from the outside that no bad actor has ever touched any market, and neither can anyone else writing about this. I can say what real evidence would look like: a settlement value that does not match the average of the 60 index prices for that window, fills at prices that never existed in the book, or a demonstrated path by which activity on Kalshi's book feeds back into the CF Benchmarks index. All three are testable, because the source, the rule and the terms are public.
I ran the simplest test on the 200 windows: in every one, the result matched the sign of Kalshi's own published settlement value against the line, with zero exceptions. That rules out clerical mis-settlement and nothing more. The real test, the settlement value against the 60 index prices from that minute, is still open, and nobody in the angry threads has run it. Rigging outcomes would mean moving the index itself, across every venue CF Benchmarks draws from, during one exact minute, which is a far bigger job than pushing prices on one exchange and would leave footprints in public price data. Kalshi is a CFTC-regulated exchange with surveillance obligations, which is why there is a legal framework around these products at all.
My Take
I do not chase the final two minutes with market orders. That is when spreads are widest and I would be trading against index-reading programs on their best terrain. If I want late exposure, I rest a limit order at my price and accept that it may not fill.
The fee schedule says the same thing. Kalshi's taker fee peaks exactly where near-the-line contracts trade (the full math is in our Kalshi fees breakdown):
| Contract Price | Taker Fee Per 100 Contracts | Fee As % Of Position Cost |
|---|---|---|
| 10 cents | $0.63 | 6.3% |
| 30 cents | $1.47 | 4.9% |
| 50 cents | $1.75 | 3.5% |
| 70 cents | $1.47 | 2.1% |
| 90 cents | $0.63 | 0.7% |
A round trip near 50 cents, in and out as a taker, costs about $3.50 per 100 contracts, roughly 7% of a $50 position before you have won anything. Scalping coin flips at the fee peak is a hard way to make a living, and it is one more reason the fast money in the last minute is a program, not a person leaning on market orders.
When I do like these markets, it is earlier in the window, at prices away from 50 cents, when I have an opinion about short-term volatility rather than a hope about the next wick. For opinions that need longer than 15 minutes, Kalshi runs slower crypto boards, and our AI panel has priced both the year-end Bitcoin bands and the timing of the next $100K print. For the full shelf of 15-minute series and a live board of current windows, see our Kalshi 15-minute markets hub.
More on this: Where Does Bitcoin End 2026? Kalshi's Price-Band Odds · How Bitcoin Price Contracts Work On Kalshi · How Low Will Bitcoin Get In August? The Floor It Hit Twice · Bitcoin 250K By 2027? Kalshi's Longshot Just Woke Up · How High Will Bitcoin Go In September 2026? Kalshi Vs. AI
Keep the sports side of your workflow just as honest.The OddsShopper odds comparison does the line shopping for you across every major book, and the arbitrage finder flags the spots where two prices disagree enough to matter. Knowing the true number before you trade is the whole edge, on an exchange or at a sportsbook.
Event contracts involve risk and are not appropriate for everyone. Availability varies by state. Nothing here is financial advice.
Prices on this page are Kalshi's book. If you also trade on Polymarket, code OS4 gets new users a $50 trading bonus on a $10 deposit — affiliate link; terms as stated by Polymarket; 18+, availability varies by state.
