What BIASMACHINE is and how the probabilities are computed

Why we do not predict the market

Price is made of events nobody can predict in advance: news, large orders, coincidences. Trading theories try to sort these events into patterns and give them names: "accumulation", "manipulation", "buying pressure". A name tells you nothing about the next move.

What can be done honestly is to count how often, after a specific event, at a specific time, on a specific asset, price went further and how often it came back. That picture does not depend on theories or definitions: there is an event, a time window, a sample and a frequency.

What an event is

An event is a measurable fact on the chart tied to time: price exceeded a session high or low (a sweep), deviated from the open by a given percentage, or made a pullback or continuation of a given size within a given number of minutes.

All times are New York time. The trading day runs from 18:00 to 18:00 the next day: the US futures evening session opens at 18:00.

Data: second-level price history since 2019–2020 (the start depends on the asset); every series is aligned to New York time with daylight-saving changes handled and checked for gaps.

How algo levels (deviation quantiles) are computed

For each 15-minute interval of the day (96 of them, the first being 18:00–18:15) we take every day in the sample and measure how far, in percent, price deviated from the interval's open and from the session open. These deviations give quantiles, the boundaries of the distribution. For example, p60 means that on 60% of the sample days the deviation stayed within that boundary. The same is computed for hourly and four-hour intervals.

An algo level is where price crosses two boundaries at once: a quantile of the deviation from the session open and a quantile of the deviation from the candle open. Level A1 is the session median (p50) plus the candle p60; level A2 is the session p70 plus the candle p80. If price reached A2, such a deviation at that time of day has been rare historically.

A level does not say "buy" or "sell". It says where price is relative to how far it usually travelled at that time. The standard sample is trading days filtered by month and weekday, without days with special conditions such as major news and futures contract rolls. In Pro you can set the sample yourself: dates, weeks, months, weekdays.

How sweep, pullback and continuation frequencies are computed

Sessions are fixed New York intervals. For each day in the sample we record whether one session swept the high or low of another and what happened after the sweep within an observation window of 15 to 180 minutes: a pullback of a given percentage from the swept level, a continuation of a given percentage, or neither. The result is the share of each outcome and the number of days in the sample.

Example from the video of 4 October 2026 (the figures will be re-checked after the Research recalculation): on US100 the 07:30–09:30 session high was swept on 594 of 848 trading days. In 70% of those cases a 0.2% pullback came before an equal move in the sweep direction, and in 28.7% neither happened within 90 minutes.

Why the same hypothesis gives different numbers on different assets

Assets distribute their intraday movement differently and react to sweeps differently. Example from the video of 24 October 2026 (the figures will be re-checked after the Research recalculation): a 0.4% pullback after a sweep of the high formed at 03:00–04:30 New York, within 09:00–15:00, occurred on 41.8% of days on US100 and on 21.7% of days on gold — under the same rule.

That is why every study in BIASMACHINE is computed separately for each asset. The platform offers no "works on any market" strategies.

What a percentage means and what it does not

A percentage is the share of historical cases that satisfy the rule in a sample of M days over the stated period. It is not a probability of profit, not a win rate and not a forecast. With a small sample of tens of cases the figure is unreliable: look at the number of cases next to the percentage.

Historical frequencies are not individual investment advice and do not guarantee repetition. Hypothetical results do not reflect real trading (CFTC Rule 4.41 principle).

How to check the calculation yourself

Every study states its rule, window, sample and period, so the calculation can be repeated on your own data.

In the free tester you can test a hypothesis by trading on history; deviation quantiles and basic statistics are available without a subscription. In Research (Pro) every case in the sample opens on the chart with one click, showing where the sweep, pullback or continuation was recorded.

What is free and what is Pro

Free:

• Tester and trading on history — regular practice attempts
• Deviation quantiles and basic statistics
• CFD 10k and futures-rules 50k challenges (the latter is labelled CME 50k in the interface; the platform is not affiliated with CME)

Pro:

• Realtime and algo levels for all assets
• Research: scenario studies and the experimental Signal Bar
• Advanced replay: date selection, higher speed, fast-forward and extra history
• Blind mode in the tester
• Other challenge account sizes and prop-firm rules, Funded continuation after a notional payout
• Advanced trade journal and analytics
• Telegram notifications

Pro prices as of 10 November 2026:

• €90 for 3 months (€30 per month)
• €252 for 1 year (€21 per month)
• €648 for 3 years (€18 per month)
• €900 for 5 years (€15 per month)

One-time payment, no auto-renewal. Pro unlocks every feature and every asset.

The platform is not a broker, takes no money under management, gives no signals or individual investment advice and sells no courses.

More on the Access page and in the FAQ.