A Kalshi BTC 15-minute market (series KXBTC15M) resolves Yes, meaning Up, if the simple average of the 60 CF Benchmarks BRTI prints in the final minute is at least the same 60-second average taken just before the window opened. Both averages are rounded to the cent and an exact tie goes to Yes. The BRTI is an index built from the order books of several large exchanges, so it is not the price on any one exchange's chart.
That difference shows up in the data. Across 18,023 officially settled windows from March 24 to October 4, 2026, a Coinbase/Kraken spot read at the close pointed the opposite way from Kalshi's result 5.06% of the time, and 38.6% of the time when spot was within $5 of the strike. The Kalshi ask itself tracked the outcome closely at both five and three minutes before the close. Weekend windows ended within 0.02% of the strike nearly twice as often as weekday windows.
Every KXBTC15M market carries its own rule text. Here is one, verbatim, from the window that closed at 7:00 PM EDT on October 4, 2026.
"If the simple average of the sixty seconds of CF Benchmarks' BRTI before 7:00 PM EDT on Oct 4, 2026 is at least the simple average of the sixty seconds of CF Benchmarks' BRTI before 6:45 PM EDT on October 4, 2026, then the market resolves to Yes."
Kalshi's market notes add: "Not all cryptocurrency price data is the same. While checking a source like Google or Coinbase may help guide your decision, the price used to determine this market is based on CF Benchmarks' corresponding Real Time Index (RTI). 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 CF Benchmarks Bitcoin Real Time Index is calculated from the order books of a set of constituent exchanges rather than from one venue's last trade. CF Benchmarks' constituent list (version 13.7, August 24, 2026) names Bitstamp, Coinbase, Kraken, Gemini, LMAX Digital, Bullish Exchange and Crypto.com for the BRTI, and shows itBit suspended from it on July 27, 2026. Kalshi collects 60 published BRTI values in the final minute, one per second, and averages them.
Two consequences follow. A single exchange can print a different price from the index at the same moment. And because settlement averages a full minute, the price in the last few seconds matters less than it looks: a late spike moves a 60-second average by a fraction of its size.
We compared the direction our own Coinbase/Kraken spot feed showed at the close with Kalshi's official result, window by window.
| Spot price at the close, distance from the strike | Windows | Spot read disagreed with Kalshi's result |
|---|---|---|
| $0 to $5 | 852 | 38.6% |
| $5 to $10 | 923 | 23.2% |
| $10 to $25 | 2,392 | 10.1% |
| $25 to $50 | 3,463 | 2.8% |
| $50 to $100 | 4,681 | 0.5% |
| $100 or more | 5,712 | 0.1% |
| All windows | 18,023 | 5.06% |
Within $5 of the strike, a spot chart was wrong more than a third of the time. Within $10 it was wrong on 543 of 1,775 windows, about three in ten. Past $50 it was almost never wrong. If you are watching a close window settle, the chart you are looking at is a weak guide to the result.
Is the Kalshi BTC 15-minute price accurate? We checked the live ask at two points in each window against how often that side actually won.
| Ask for the side | ~5 min left: windows | won, at avg ask | ~3 min left: windows | won, at avg ask |
|---|---|---|---|---|
| 1 to 20¢ | 185 | 11.4% at 14.8¢ | 412 | 9.7% at 12.2¢ |
| 21 to 40¢ | 1,023 | 32.5% at 32.3¢ | 773 | 30.4% at 31.2¢ |
| 41 to 59¢ | 1,901 | 50.1% at 50.7¢ | 1,136 | 49.1% at 50.6¢ |
| 60 to 69¢ | 1,497 | 63.1% at 64.8¢ | 826 | 61.4% at 64.8¢ |
| 70 to 79¢ | 2,098 | 74.2% at 74.7¢ | 1,135 | 73.1% at 74.9¢ |
| 80 to 89¢ | 3,202 | 84.7% at 84.9¢ | 1,823 | 84.4% at 84.9¢ |
| 90 to 94¢ | 2,982 | 93.0% at 92.1¢ | 2,107 | 91.2% at 92.2¢ |
| 95 to 97¢ | 2,663 | 96.5% at 96.1¢ | 2,320 | 96.3% at 96.2¢ |
| 98 to 99¢ | 2,317 | 98.4% at 98.5¢ | 4,558 | 98.5% at 98.6¢ |
In most rows the realized rate sits within a point or two of the ask, and usually a little under it, consistent with the spread a buyer pays. That is what a well-priced market looks like. The squared-error score of the ask as a forecast improved from 0.1225 at five minutes to 0.0963 at three minutes, as the outcome became more settled. One disclosure: the five-minute file stored a wrong win flag on 3,699 rows (it copied the later signal's result). Every rate above is recomputed from the logged side and Kalshi's official outcome.
A window that ends within 0.02% of the strike, about $14 at $72,000, is the one most exposed to Finding 1. How often that happens depends on when you look.
| When the window closed | Windows | Ended within 0.02% of strike | Median move |
|---|---|---|---|
| Weekday | 12,756 | 10.7% | 0.101% |
| Weekend | 5,267 | 20.2% | 0.059% |
| Midnight to 6 AM ET | 4,270 | 15.0% | 0.077% |
| 6 to 9 AM ET | 2,288 | 14.7% | 0.078% |
| 9 AM to noon ET | 2,289 | 10.0% | 0.128% |
| Noon to 4 PM ET | 3,058 | 12.3% | 0.093% |
| 4 to 8 PM ET | 3,060 | 14.8% | 0.080% |
| 8 PM to midnight ET | 3,058 | 12.9% | 0.087% |
| Calm hour (previous hour's moves in the lowest third) | 5,793 | 20.4% | |
| Normal hour | 5,794 | 12.0% | |
| Volatile hour (highest third) | 5,794 | 7.9% | |
| All windows | 18,023 | 13.5% | 0.087% |
Weekends and calm hours had nearly twice the share of knife-edge windows. The US morning session, 9 AM to noon ET, has the largest moves and the fewest knife-edge closes.
Near-flat windows are also visible before the close. When Bitcoin sat within 0.02% of the strike about three minutes out, 2,633 windows had a usable ask: the ask for the side the move pointed to averaged 52.1¢, and that side won 50.7% (95% range 48.8 to 52.6). The market priced those windows as coin flips, and they were. Our published feed rarely fires on them: only 14 of its 11,537 graded signals came from such windows.
The regime changes the win rate, and the price moves with it. Our published record won 93.2% at a 93.2¢ average entry in calm hours, 90.6% at 91.8¢ in normal hours and 90.2% at 90.5¢ in volatile hours. On weekends it won 0.5 points less often than its 93.3¢ average entry, and on weekdays 0.6 points less often than its 91.2¢ entry. Calm, high win rate, high price; busy, lower win rate, lower price. No slice beat its own price by more than chance. One time slice, 4 to 8 PM ET, won 90.5% at a 92.2¢ average entry, below its price by 2.78 standard errors. That clears the bar for testing six time slices (2.64) but not the bar for every slice we report across these studies (3.04), so it may be noise.
When Bitcoin is well clear of the strike with three minutes left, there is frequently nothing left to buy.
Of the 3,152 windows where Bitcoin was 0.20% or more from the strike about three minutes before close, at least 2,032 (64%) had no ask below $1.00 on the side Bitcoin was on. Across our whole log, 3,099 graded signals were priced at 100¢ and are excluded from our published record for exactly that reason: a contract that costs a dollar and pays a dollar has no trade in it. A strong late move is not an opportunity that went unpriced. It is usually an opportunity that is already gone.
A measurement, not a recommendation.
Related: what 9,800 graded Kalshi BTC 15-minute windows show, Polymarket 5-minute vs Kalshi 15-minute, and what a graded signal is.
Per Kalshi's market rules, checked October 4, 2026, a KXBTC15M market resolves Yes, meaning Up, if the simple average of the 60 CF Benchmarks BRTI values in the final minute before the close is at least the same 60-second average taken before the window opened. Both averages are rounded to the cent.
The CME CF Bitcoin Real Time Index, published by CF Benchmarks. It is built from the order books of several constituent exchanges, listed in August 2026 as Bitstamp, Coinbase, Kraken, Gemini, LMAX Digital, Bullish Exchange and Crypto.com, rather than from any single exchange's last trade.
It resolves Yes. The market's strike type is "greater or equal", so a settlement value equal to the strike counts as Up.
Kalshi settles on a 60-second average of the BRTI, not on a Coinbase tick. In our data a Coinbase/Kraken spot read at the close pointed the opposite way from Kalshi's official result on 5.06% of 18,023 windows, and on 38.6% of windows where spot was within $5 of the strike.
In our data, yes. Grouped by price, the live ask about five minutes and about three minutes before the close sat within a point or two of how often that side actually won in most groups, usually a little above it, which is the spread a buyer pays.
On weekends and in calm hours. 20.2% of weekend windows ended within 0.02% of the strike against 10.7% on weekdays, and 20.4% of windows after a calm hour against 7.9% after a volatile one. When Bitcoin was that close to the strike three minutes out, the ask averaged about 52¢ and that side won 50.7% of the time.
Kalshi lists trading as 24/7 apart from scheduled maintenance on Thursdays from 3 to 5 AM ET. Our sample includes 5,267 weekend windows.
How each number was produced.
Our logger records one observation per KXBTC15M window, alerted or not. This page uses the 18,025 windows where Kalshi's official result was read back from its public API within about 30 seconds of the close; two rows with a corrupted strike (logged near $7 on April 13, 2026) are dropped wherever a strike is needed, leaving 18,023.
The ask is one minus the best opposing bid in Kalshi's live order book, rounded to the cent, for the side our model favored. In the calibration table "won" means that side matched Kalshi's official result.
Regime win rates use our published BTC record (11,537 graded signals, the same set as the results log); each rate is shown with the average price paid in the same slice.
Venue rules as published in each source, checked October 4, 2026.
Citations welcome; the CSV above is the primary source and this page is the canonical writeup. Past performance does not guarantee future results.
The free, verifiable part comes first.
Every signal in our BTC 15-minute feed posts to a free public channel and a public results log, graded against Kalshi's official result with nothing pulled.
This is a free trial that becomes an automatically renewing subscription. A payment card is required to start, and you are not charged today. Your trial runs 30 days. At the end of it your card is charged $4.99 and then $4.99 every month until you cancel. Cancel any time before the trial ends and you pay nothing: use your account page, send /manage in the bot, or email [email protected]. Cancelling later stops future renewals at the end of the current paid month and does not refund it. No refunds except as required by law. By starting a trial you authorize these recurring charges and agree to the Terms and Privacy Policy. Signals are informational and for entertainment only, are not financial or betting advice, and any single signal can lose 100% of your stake. Must be 18+ (some platforms require 21+). If gambling is a problem for you, call 1-800-GAMBLER.
These are 15-minute binary prediction-market contracts. Any single window can lose 100% of your stake. They settle on a 60-second index average, not on the exchange chart you may be watching, and close windows can settle against what your chart shows. Nothing on this page is financial, investment, or betting advice.
Past performance does not guarantee future results. Trade only with money you can afford to lose. Must be 18+ or of legal age where you live. If gambling is a problem for you, call 1-800-GAMBLER.