A round of four to six minutes against a chart that was generated rather than recorded. Three series arrive one at a time, each with 22 bars of history and a band drawn around the last close; you get eight calls per chart and ten seconds each, and one tap says both things at once — up or down, and inside the band or past it. The twist is in the scoring. Direction pays nothing, because in this simulation the next bar's direction is drawn independently and the best possible strategy is "always up", whose ceiling is the up-day share this simulation draws at. Size does pay, because quiet bars follow quiet bars and violent ones follow violent ones, so the last three bars are real information. The result card puts your direction count beside a coin and beside an always-up strategy, runs a direction test that reports an e-value rather than a verdict about you, and then grades you on size alone.
Three charts a round, 24 calls in total. Each chart opens on a reading screen with 22 bars of history, a baseline at the last close and a band above and below it labelled as a percentage; you start the clock yourself. Then come eight bars, ten seconds each, and after each one the new candle is drawn and judged. Nothing ends the round early: there are no lives, and a wrong call costs the streak and nothing else. A round takes four to six minutes.
Tap one of the four buttons — small up, big up, small down, big down — and that single tap is both answers. On a keyboard, the arrow keys move inside the 2×2 grid and Enter answers, or 1 to 4 answer directly. There is no undo and no pause while a bar is live, because the clock is the pressure; the reading screen between charts and the recap after each chart both stop the clock. A bar where both halves were right moves on by itself after 2.8 seconds; any bar with an explanation waits for "next bar", so there is time to read it.
Size is worth 60 points a bar, and the multiplier rises half a step every two correct sizes in a row up to 2.5×. Direction is worth zero — not a smaller number, zero — and the streak ignores it entirely. A wrong size resets the streak; a wrong direction does not. Running out of time counts as missing both halves, costs the streak, and costs no points. The grade reads the share of the 24 bars whose size you got right, so the grade cannot be improved by direction luck.
The round ends after the 24th bar. The result card opens with the direction block — you, a coin and an always-up strategy, all out of 24 and all marked as diagnostic — then the direction test: the number of bars you answered, how many you got right, and an e-value against a threshold of 20. Crossing that threshold carries a false-alarm bound of 5%; not crossing it means the round gives no reason to think direction reading was at work. Below that sits the size line against random guessing (about 12 of 24), the grade with a note that it reads size only, three knowledge cards, and a bar-by-bar review that keeps the sentence each wrong bar was given.
Every round is dealt from one seed and the share link carries it as ?run=seed.version: opening that link deals the same three charts in the same order, in either language, because the deal is language-neutral. A link from an older version still opens and says the round was rebuilt with the current one. "Play again" draws a new seed. Leaving mid-round keeps the bar you were on with the score, the streak and the seconds that were left; the share text and the saved best record carry the size score and the grade only, never the direction count.
Both languages play the identical round — the same seeds, the same bars, the same bands — and only the sentences differ, because nothing in the generator depends on where you live. The one real-world figure in the game, the 53.7% of S&P 500 days that closed up between 1951 and 2025, is named with its index and its period in both languages rather than being presented as a universal rule, and the knowledge cards are written separately in each language rather than translated.
In the S&P 500, 53.7% of trading days between 1951 and 2025 closed above the previous close, ranging from 51.3% to 54.9% depending on the decade. That tilt is small, it is the entire edge an always-up strategy has, and it is the rate this game's generator draws at.
Source: S&P Dow Jones Indices daily close history, 1951-2025
Across markets and periods, the autocorrelation of returns is close to zero while the autocorrelation of absolute returns stays positive for many lags — quiet stretches follow quiet stretches. That is why size is the half of this game that can be read, and the careful statement about real markets is narrower than "direction is random": a few bars are rarely enough to call the next day.
Source: Cont 2001, Quantitative Finance 1(2), 223-236, stylized facts
Published work from 1959 generated price paths from random draws and showed that they produce the shapes chart readers name and trade on. Every series in this game is generated the same way, which is exactly why none of them is given an instrument's name.
Source: Roberts 1959, Journal of Finance 14(1), 1-10
The direction test multiplies a fixed payoff for each call you made, which gives a quantity whose expected value is 1 when direction choices carry no information. Crossing 20 therefore happens at most 5% of the time by chance. That is a bound on false alarms, and it says nothing about how likely luck was in your particular round.
Source: Ville 1939 inequality; review in Statistical Science 38(4), 2023
A study of tens of thousands of household brokerage accounts found that the most active fifth trailed the market badly after costs, while the least active came close to it. A game about calling the next bar is worth playing precisely because it shows how little direction reading is worth.
Source: Barber & Odean 2000, Journal of Finance 55(2), 773-806
Because in this game it cannot be read. The direction of each bar is drawn independently of the regime and of every earlier bar, so conditioning on the last one to three bars moves the up rate by at most about one percentage point, and the best possible strategy is "always up", and its ceiling is this simulation's own up-day rate. Paying for direction would mean paying for a coin flip: an earlier draft awarded 40 points a bar for it, which put up to 31% of a perfect score inside luck. Direction is still counted and shown, beside a coin and beside always-up, as a diagnosis.
Not predicted, but it can be read. Big moves arrive in runs, so following the previous bar's size hits about 67% of the asked bars in this game, while always answering "small" hits about 48%. That gap is the game. It is also the one property of real markets this simulation is built on: the autocorrelation of absolute returns stays positive for many lags, while the autocorrelation of returns themselves sits near zero.
It is a betting score. For each bar you answered, the test multiplies by a fixed factor that pays more for a correct down call than a correct up call, chosen so that the running product has an expected value of 1 if your direction choices carry no information. If it ever reaches 20, that happens at most 5% of the time by chance — whatever you based your calls on and however many bars the clock took. So the card reports the e-value, the threshold and that bound, and deliberately nothing about how likely luck was in your own round — the number does not carry that meaning.
No — every bar is generated by the game from one seed rather than recorded from any market. A share link reproduces your exact round for the same reason. There is no instrument, no ticker, no currency and no money anywhere in the game, and the series are named A, B and C rather than given an asset's name. The one real-world figure in the game is the share of up days quoted with its index and its period on the result card.
Yes — the bar you were on, the score, the streak and the seconds left are kept on your own device, and reopening the game resumes that bar. What is kept is the list of taps rather than a score, so a resumed round cannot disagree with the answers that produced it. The best record and the share text carry the size score and the grade only.
No, and the result is designed to show why. The direction half is a coin flip by construction, the size half rewards noticing that volatility comes in runs, and nothing in the game involves money, positions or profit and loss. The reading it does train is a narrow one: how large a move is, relative to how large recent moves have been.
No investment advice, forecasts or recommendations. No real security is named and no account can be opened here.
