Questions, answered

The ones we call vetted are ahead over 100 or more markets after we found them, by more than luck explains: their returns are at least 3.5 standard errors above what breaking even would give at the prices they paid, on markets that finding them can't flatter. The bar comes from a simulation of the leaderboard itself, which looks again after every market: a signal that only breaks even gets vetted in its first month about 1 time in 100, so with dozens being tested an occasional lucky one can still get through. Their records are live on the Lab, the home page and the Accuracy page, wins and losses; signals taken off are listed on the Lab with their records. A new find starts on trial and shows how far it has to go ("62% vetted": the further-off of its 100 markets and its 3.5 standard errors). The dollar figures are hypothetical: the same stake the first time a signal fires in a market, at the asking price with the fee.

That is what it is built for, and it is scored on exactly that, live. The model starts from the market's own price and moves off it only where 14 days of recorded markets say the price has been wrong: 33 live readings, from order flow on 11 exchanges and the fastest futures to the other coins' markets, refitted every hour so what stops working is dropped. When its line pulls away from the price, the Live page shows the side with the edge, at what price, after the fee, and its buy signal fires when that edge reaches 3 points. Every lean is logged and scored against what the price said: the Accuracy page shows how often its side won, wins and losses. So far it is not ahead: in its first two days its side won less often than the price said, so it now leans only where a test on markets it was never fitted on shows it beating the price by more than luck (two standard errors); elsewhere it is the market's price. Our first model, built from Bitcoin's price alone, tied the market; that test stays published too.

Three things. Signals that are tested, then vetted: the Lab tests hundreds of thousands of signals every few minutes and keeps only the ones that keep winning on markets they were not found on. A public record: every market is scored on the Accuracy page, the SettleMath model's leans and every signal, wins and losses; most callers never show theirs. Tools to find your own edge: the strategy simulator, the penny tester, whale orders from 11 exchanges, 33 live readings, and signals you build yourself, on the chart and as notifications on your phone.

Every 15 minutes a new window opens. Its strike is the official settlement value of the window before. At the close, the official value is the average of the index's 60 one-second values before the close, not the price at the closing second. If that average is at least the strike, Yes wins; below it, No wins. The Ethereum market works the same way on Ethereum's index.

Because in the final minute the price doesn't decide anything on its own. An example: Bitcoin is $40 above the strike with 20 seconds left, so a price chart says Yes looks comfortable. But the first 40 seconds of the final minute averaged $30 below the strike. If Bitcoin holds where it is, the 60-second average comes to about $7 below the strike: No. For Yes, the last 20 seconds would have to average $60 above the strike.

Most price tools show only the price, so in the final minute they can point the wrong way. The settlement panel averages the seconds already locked, shows how many there are, and works out what the rest must average for the result to change. We also measure how far our own price runs from the official value, every window, and build that into the model, so near the strike it says too close to call instead of guessing.

Once a second of the final minute has passed, its price is part of the average for good. With 30 seconds left, half the average is fixed; with 10 seconds left, five sixths. The fewer seconds remain, the further the price has to move to change the result, and the panel shows exactly how far.

Our price is our own composite of three exchanges, not the official index, so the two differ by a few dollars. When the average lands within that measured difference of the strike (the 95th percentile over the last 7 days), our price alone can't say which side the official value falls on, and the page says so instead of naming a side.

From the public feeds of exchanges that are part of the official index. The price on the chart and the Live page merges five of their order books (Coinbase, Kraken, Bitstamp, Gemini and Crypto.com) and takes a depth-weighted mid, the way the index is built; over its first 299 markets its final-minute average missed the official value by a median $3, against $5 for our older price, the median of three exchanges' best prices, which still drives the model and the signals for now. The official index itself needs a license to use in a product, so we don't redistribute it. How far our price has been from the official value is measured every window and shown on the Accuracy page.

Its chance that the market ends Up: the market's own price, moved by what the 33 live readings have been worth against that price over the last 14 days (a ridge-penalized logistic fit, refitted every hour, its penalty chosen on markets it was not fitted on; in the last minute it can also weigh the seconds already averaged). Each part of the market, and that last-minute arithmetic, is used only when it has beaten the price by more than luck (two standard errors) on markets it was never fitted on; otherwise the model is the market's price there. Its move off the price is drawn twice as large on the chart's line, so a lean is easy to see; the number, the edge and the record use the chance itself. Where the readings say nothing, it sits on the market's price. Its record is on the Accuracy page.

Where the official settlement value is expected to land: the same model's spread, read as prices instead of a chance. The solid part is the likely range (a 50% chance it lands inside), the faint part the wide one (90%). It narrows as the final minute locks in. How often the official value actually landed inside each range is scored on the Accuracy page. Pro.

Single market orders of 10 BTC or more on any of six spot exchanges, or 25 BTC or more on five perpetual futures venues (where orders run far larger), shown the second they happen and marked on the chart: green triangles where the taker bought, red where the taker sold. Exchanges report an order as its separate fills, so trades on one exchange and one side are counted as one order when each comes within a quarter of a second of the last and all within half a second; a stream of small orders never adds up to a whale. A big order can move the price for a moment; whether it moves a 15-minute result is another matter, so they are context, not a forecast. Pro.

33 live readings, each scored on its own record and all feeding the SettleMath model. Fifteen carry points: RSI, Bollinger %B, MACD and a short trend (EMA 9 against EMA 21) on one-minute closes of our price; the 1- and 5-minute change; the share of the last minute's volume that takers bought on 11 exchanges; which side the order books of 5 exchanges lean to; how fast the futures premium is moving; the Coinbase premium (Coinbase's price against Binance's) over 5 minutes; the whale orders of the last 5 minutes; the market's own book, its price over the last minute, and which side its own trades are taking (the crowd); and how the other 15-minute Bitcoin up/down market prices the same window against this one. Watched and scored too, with no points until their record shows they help: forced liquidations on the futures exchanges, Bitcoin against the window's volume-weighted average price (VWAP), new positions behind the move (open interest), the options' skew, whether spot or the futures led the move, how often history says the market's price ended Yes, the market's own big trades, streaks of windows, orders pulled from the market's book, whether the ETH market moved first, the market's own price velocity, where Bitcoin sits in its 5-minute range, whether the move is accelerating, the fastest exchanges moving ahead of our price, the other coins' markets moving first, and the final-minute arithmetic against the price. Each points Up, Down or neither and carries points, 100 between them: every one pointing Up adds its points, every one pointing Down takes them away, so the strength score runs from 100 Down to 100 Up. The points are starting values; the Accuracy page scores every indicator on its own, and those records decide which ones earn more. None of them is a forecast. Pro.

Which side the people trading this window are taking right now: of the market's own trades in the last 2 minutes, the share whose buyer took Yes, one trade one vote whatever its size, so a crowd of small traders counts as much as a few big ones. It is drawn as a dotted line on the chart's 0-100 scale and as a small chart beside it, and it is one of the 15 indicators, scored on the Accuracy page like the rest. Pro.

SettleMath model: once a market, with 14:00 to 0:30 left, the model's chance of a side sits 3 points or more above that side's asking price after the taker fee (between 5¢ and 95¢). A line on the chart marked Buy Up or Buy Down, and a row on the leaderboard, where it is vetted like every other signal: ahead by more than luck explains over 100+ markets.

Lock: In the last minute (1:00 to 0:08 left), our settlement math - the seconds of the 60-second average already in and where Bitcoin is now, without the market's price - puts one side at 97.5% or more while the market has it at 85¢ to 99.5¢, or at 99% or more above that, up to 99.8¢. Each win pays a few cents and a rare loss costs many wins. (Lock v6, from Oct 28, 3:38 AM CT; v5 needed 98% at 90-99¢ and fired in about one market in five. Earlier versions are on the changes list, with their records as a download.) (Indicators agree was retired on Oct 24: when the indicators line up strongly the market has almost always already priced that side, so it paid no better than the price; its records stay on the Accuracy page.) It is a line on the chart and a notification on your phone, and every one is logged: the Accuracy page shows how often the side it pointed to won, next to what the market was already pricing. They describe what the numbers did; they are not advice. Pro.

Yes. Combine up to six conditions on the page's numbers (the market's price, the model, the flip chance, RSI, volume, order flow, time left and more), pick a color, and a line is drawn on the chart every time they all line up. Signed in, your signals are kept on your account with their record across windows, and you can have them sent to your phone as notifications. Pro.

Every window since our collector started is scored on the Accuracy page, live, never a backtest: the SettleMath model's leans against what the price said, every signal's wins and losses, our first model against the market's own price by time left, and our price against the official value. The market's price is a strong forecaster and we publish every comparison, including where it does better. What Pro gives you is the model, the signals and the tools live, not a promise about results.

The free view is the whole Live page, 60 seconds behind, with some panels locked. In a 15-minute market a minute is a long time: in the final minute, the free view shows the settlement locking in after it has already happened. Pro is the same page live, with whale orders, indicators, signals, the flip chance, base rates, the order book and the full history tables. See the full list.

No. The free view (the whole Live page, 60 seconds behind) is open to everyone, and the 24-hour pass ($4.99) gives Pro for a day. The pass is not a subscription: it is charged once and never renews, so there is nothing to cancel.

Yes, from your account page in two clicks. Pro runs to the end of the period you paid for, and nothing more is charged.

Payments run on Stripe, a PCI Level 1 certified payment processor. You enter your card on Stripe's own checkout page; SettleMath never sees your card number and stores no card details. Your account here holds your email and a Stripe customer reference.

One payment of $299 for Pro for as long as SettleMath runs: no renewals, every new feature and coin page included, your Lifetime number and a Lifetime role in our Discord, early access to new features, and higher limits (30 of your own signals, longer Lab scans, notifications on more browsers). Only the first 73 people get it; the Pricing page shows how many spots are left. It is not refundable, except a full refund if SettleMath shuts down within 12 months of your purchase, or for a problem caused by a bug on our side (see Refunds).

Yes: the 15-minute Ethereum market has its own Live page (the ETH tab at the top), with the same model, the settlement panel. The trade-history panels, whale orders, indicators and signals are Bitcoin only for now.

Yes: the lock and indicator signals, and your own signals, as a notification on your phone or computer the moment they happen, even with the page closed. One switch, on the Live page's Signals card or the Signals page. On an iPhone, add SettleMath to your Home Screen first (Share, then Add to Home Screen) and turn them on from there. Pro.

Yes. On an iPhone or iPad, open the site in Safari, tap Share, then Add to Home Screen; on Android, open the browser's menu and tap Add to Home screen or Install app. It then opens full screen from its own icon, like an app, and on an iPhone that is also what lets it show notifications.

No. It is an information site. It never places orders, never connects to anyone's trading account and never holds funds. You trade, if you trade, on your own account elsewhere.

No and no. The numbers are information about the market and our model, with their record shown in full. Event contracts involve risk of loss, and past accuracy does not guarantee future accuracy. See the Disclaimer.

No. SettleMath is independent and is not affiliated with, endorsed by or sponsored by the exchange that lists these contracts, CF Benchmarks, or any exchange or data provider named on the site.

The live numbers update every second, and each carries a badge: green when its data is under 3 seconds old, amber to 10, red after, so a stalled feed is never mistaken for a quiet market. The free view's numbers are 60 seconds old by design.