Chart or Noise

How it works, exactly

The round

• Ten pairs a day, the same ten for everyone, new at 00:00 UTC.
• One chart of each pair is real: 200 consecutive four-hour candles (33 days and 8 hours) of one market. The other is noise made from those same candles (see below).
• Both are drawn by the same code, on the same vertical scale, starting from 100. No market name, no dates, no prices, no volume. Which one is A and which is B is drawn at random.
• Eight seconds a pair. No tap in time is a miss. Leaving or reloading while a pair is on screen also counts as no tap.
• After each answer you see which was real, the market and the dates.
• Your first run of the day in this browser is the one that counts. It is kept in this browser only: there is no account.

How the noise is made

1. Take the 200 real candles and the close of the candle just before them. That close is the starting price.
2. Write every candle as four ratios to the close before it: its open, high, low and close.
3. Put the 200 candles in a random order (a Fisher–Yates shuffle, every order equally likely, drawn on the server).
4. Rebuild a chart from the starting price by applying the candles one after another in the new order. Multiplying does not depend on order, so the noise ends on exactly the price where the real chart ends.
5. Round every price the way the real prices are rounded: to the most significant figures and the most decimals found among the real window's prices. Both charts then sit on the same price grid.

The real chart goes through the same steps without the shuffle and comes out as the real prices. So the two charts hold the same 200 candles, start at the same price and end at the same price. The only difference is the order.

How hard this is

Hard. Guessing blind averages 5 out of 10, and most first tries will not land far above that. Trends, ranges, breakouts and every pattern with a name show up in both charts, because shuffled candles draw them too.

What the shuffle destroys is order. In a real market big candles tend to come in bunches and quiet candles follow quiet candles; in the noise the big candles are sprinkled evenly. That difference is measurable: a small program that looks at nothing else (how well the size of one candle predicts the size of the next five) picked the real chart in 1,455 of 1,658 test pairs we made this way from the same twenty markets, 88%. Whether eyes can find it in eight seconds is the open question, and the crowd average is the answer: the mean score of all rounds finished today. With only a few rounds in, it means little; the page always shows how many it is made of.

The markets and the windows

The real charts come from twenty perpetual markets on Hyperliquid: the twenty with the largest traded volume over the 30 days to 11 November 2026 (each day's volume times its close, summed; measured on 12 November 2026 among the 45 markets with the largest 24-hour volume that day). BTC, ETH, HYPE, ZEC, SOL, NEAR, PUMP, XRP, XMR, kPEPE (1,000 PEPE), ENA, WLD, LIT, UNI, BNB, ZRO, SUI, XPL, VVV, AAVE. All are crypto markets from the exchange's own list; stock and pre-listing markets added by outside deployers are not in it. The list is fixed; it does not re-rank itself.

Each day ten different markets are drawn, and for each one a window of 200 four-hour candles that starts on a candle boundary. The exchange serves only the most recent 5,000 candles of a timeframe, so a window starts at most 4,800 candles (800 days) before the day of play, or at the market's first candles where it is younger than that (HYPE, ZEC, PUMP, XMR, LIT, XPL and VVV have fewer candles there, so their windows repeat sooner), and ends at least 24 hours before the day of play. No two windows of a day come within 24 hours of each other. A window is used only if all 200 candles exist and at least 150 of them moved.

Candles are read from Hyperliquid's public API. Chart or Noise is not affiliated with Hyperliquid.

What a score is worth

Nothing but the fun of it. Which chart is real is not in this page or in anything it downloads before you answer: the server keeps the day's draw and answers one pair per tap. But there are no accounts, so "first run counts" is kept only by your browser, and someone who asks the server directly, or plays twice, can know the answers. There are no prizes. When you finish, the page sends your ten picks; the server scores them and adds one round to the day's count. That count can miss rounds that finish at the same instant.

The token: Chart or Noise ($CHARTNOISE)

The game is free and does not use the token. It never asks for a wallet.

Treasury wallet

See the treasury on Solscan

Token address (mint)

Launching soon