Blind Bitcoin Chart Prediction Game Scored with the Brier Score

The Prisme Arena is a free game, with no account, that tests how well you read a Bitcoin chart blind: 48 hours of bars with no date, price rebased to 100, up or not up, confidence from 50 to 99%. The answer is locked, then scored with the Brier score, BS = (1/N) Σ (f − o)².

Play a round. Bitcoin only. Exploration, not a verdict.

Published 17 October 2026. Research on past data, not investment advice.

The protocol of a round

During a round you have the workshop's tools on the visible past only: 15-minute, 1h and 4h timeframes, log scale, six indicators (EMA, SMA, RSI, MACD, Bollinger, ATR), levels, trend lines, zones and measures. The 1h and 4h buckets are aligned on the exercise's relative axis, not on real clock time, so they do not give the date away.

No server response before the reveal contains the date, the position in the archive or the window's fingerprint. If the archive changed in the meantime, the round becomes "invalid": it is never scored on other data.

The Brier score

The Brier score was proposed in 1950 by meteorologist Glenn W. Brier to score rain forecasts stated as probabilities. For one round:

score = (p − o)²

where p is the probability of a rise you stated and o is 1 if price rose, 0 otherwise. If you answer "not up" at 70%, your probability of a rise is p = 0.30. Over N revealed rounds:

BS = (1/N) Σ (f − o)²

Lower is better: 0 is perfect, 1 is the worst possible.

The slope is asymmetric. Going from 70% to 99% gains 0.09 when you are right and loses 0.49 when you are wrong. High confidence only pays if it is earned often: that is what makes the Brier score hard to game by always claiming certainty.

The two references

A score alone says nothing. The Arena shows it next to two benchmarks, per horizon:

• the 50% reference: always answering 50% gives exactly 0.250, whatever happens. Doing worse than 0.250 over time means losing information compared with "I don't know";
• the visible-past reference: the frequency of rises at that horizon within the 192 bars shown, used as the probability (absent when the visible past is too short for the horizon). It measures what you would get by stating that frequency mechanically, without reading the chart.

Beating both over many rounds is the goal. Beating them over five rounds proves nothing.

Calibration: what the score really measures

Being calibrated means being right about as often as you say. If, across all your rounds stated at 70%, price rises in 7 out of 10, you are calibrated at that level. If it rises in 5 out of 10, you are overconfident.

The Arena's contract provides, in the per-horizon summary, ten confidence bins of width 0.1 to read that gap, alongside the number of rounds per state, the ties and the mean Brier of revealed rounds only. Open, submitted or abandoned rounds never enter the score.

A trader practising in the Arena mainly learns one thing: the confidence level at which their chart reading stops carrying information.

Why this is not a ranking

Rounds overlap: with 192 visible bars and a 7-day horizon, two rounds drawn a few days apart share part of their history. A hundred rounds are therefore not a hundred independent experiments. The Arena shows a greedy non-overlapping count next to the number of rounds, and computes no independent confidence interval, no ranking and no "talent", even at 100 rounds.

Other limits, stated up front:

• a shape can be recognised despite the rebasing to 100 and the hidden date, especially around well-known large moves;
• Bitcoin only, a single source (Binance BTCUSDT).

Prisme Arena or TradingView Bar Replay?

Bar Replay is for replaying your usual chart and tools. The Arena is for finding out whether your reading carries information once the date, the price and the urge to "redo" are removed. They are different exercises: the Arena complements the work you do in TradingView, it does not replace it.

To measure in the open, without the blindfold, the levels and double tops you think you see, there is the Prisme workshop, described in measuring support, resistance and double tops without look-ahead bias.

Session, score card and privacy

• An anonymous session starts at the first "Open a round": random cookie, no account, no IP address stored. Your forecasts are never shown to anyone else.
• 500 rounds kept per session, with no deletion of bad results to make room.
• The score card is published only if you create it. It contains, per horizon, the number of revealed rounds, the mean Brier, the two references, the non-overlapping count and the ties. No individual forecast, no date, no price.

From a reading to a rule

Good calibration in the Arena does not make a strategy. If a reading keeps coming back and seems to hold, the next step is to write it as a rule and judge it on a sealed period with a trial counter: that is the job of Strategy Arena Lab, free on Windows and Linux.

FAQ

Is there a game to test how well I read a Bitcoin chart blind?

Yes. The Prisme Arena, free and with no account at strategyarena.io/prisme/arene, shows 192 BTC 15-minute bars (48 hours) with no date and rebased to 100. You call up or not up with a confidence of 50 to 99%, the answer is locked, then what came next is revealed and scored with the Brier score.

How is the Brier score calculated?

For one round it is (p − o)², where p is the probability of a rise you stated and o is 1 if price rose, 0 otherwise. Over N rounds, BS = (1/N) Σ (f − o)². It ranges from 0 (perfect) to 1; always answering 50% gives 0.25.

What is a good Brier score?

One that stays below 0.25, the score of a 50% answer, over a large number of rounds. The Arena also shows a second reference based on the frequency of rises in the visible past. Over a handful of rounds, a good score can be luck.

How is this different from TradingView's Bar Replay?

Bar Replay replays a chart you choose, with its real dates and prices; it is for practising your method. The Arena draws the window at random, hides date and price, requires a probability locked before the reveal and scores the series. The two complement each other.

Why does an unchanged close count as 'not up'?

The question is binary: will price be strictly higher at the horizon? An unchanged close is not a rise. The Arena counts these ties separately so they stay visible.

Does my score prove I can read a chart?

No. Rounds overlap in time and are not independent observations, and a shape can be recognised. The Arena publishes no ranking and no talent diagnosis: it is a calibration exercise, exploration, not a verdict.

Sources

• Prisme contracts: local Arena (edition 18: protocol, brier.binary.v1, references, calibration bins), Binance BTCUSDT archive (edition 21), public Arena (edition 22), tools in the Arena (edition 23).
• Online Arena page, read on 17 October 2026 (examples 0.09 and 0.49 at 70%, 0.250 reference).
• Glenn W. Brier, "Verification of forecasts expressed in terms of probability", Monthly Weather Review, 1950.

Exploration, not a verdict. Research on past data, not investment advice.