AI Trading Lab
Anyone can produce a profitable backtest. The Lab exists to find out whether a result is an edge or an artefact — before it reaches a customer’s account. Every X-STOX system goes through the same eleven stages, from a written hypothesis to a Myfxbook-verified record.
Machine learning does the heavy lifting in research. What ships is a set of fixed, tested rules.
Diagram · each dot is an idea, not a real count
The workflow
Each stage hands the next one a piece of evidence, and each one is a gate: an idea that fails it goes back, or goes no further. Most ideas stop somewhere along this line. That is what the Lab is for.
Research
Every system starts as one testable statement about market behaviour: what should happen, in which market, when, and why. It is written down before any data is touched, so the data cannot write it.
Historical data goes through a Python research pipeline. A classifier labels the market regime of every period, past news releases are measured, and the effect is compared with what chance alone produces on the same data.
AI at this stageMarket-regime classificationNews & event analysis
Build
The surviving idea becomes rules on paper: entries, exits, risk, behaviour around news, what to do when a broker misbehaves, and the things the system must never do.
The specification is coded as an Expert Advisor in MQL5 — MQL4 where a tool runs on MetaTrader 4 — with deterministic logic and defensive handling of what real brokers do: requotes, spread spikes, rejected orders, lost connections.
Stress
Tick by tick, with spread, commission, swap and slippage at the levels a retail broker actually delivers — including the spread widening at session boundaries and around news.
Settings are fitted on one window of history and tested on the next, unseen one — again and again, moving forward through time. An in-sample result only proves that the optimiser worked.
AI at this stageOptimisation & robustness
Results are broken out by volatility regime, trend state and session. Then the trade sequence is reshuffled, over and over, to find the drawdown the historical curve happened to avoid. The question is not what the backtest did, but what the same trades could have done in a less fortunate order.
AI at this stageMarket-regime classification
Prove
The system runs in real time on a demo account, on market data that did not exist when it was built.
Real money at small size, on a real broker: real fills, real slippage, real server behaviour — the things a demo account hides.
Release
The system ships with three risk presets — Defensive, Moderate and Extreme. Same rules, different amounts of risk, set with the stress-test results in hand, so a buyer starts from a tested setting rather than a guess.
After release the system keeps running on accounts Myfxbook verifies — track record and trading privileges — in public. A drift away from its tested behaviour sends it back to the Lab.
Stage 01 of 11· Hypothesis
ResearchWhere the AI is
Machine learning does the heavy work of research: it labels market regimes, searches and stress-tests settings, and measures how markets behave around news. None of it ships. What ships is a written set of rules, coded in MQL5, that does the same thing every time it sees the same market.
The same prices produce the same decision. Nothing inside your account is learning, drifting or changing its mind.
Every behaviour is in the specification before it is coded, and described in the manual you receive with the product.
The workflow ends on accounts Myfxbook verifies. The verified accounts are listed on the Performance page.
Each answers a research question faster, or more honestly, than a person reading charts could. Each one ends as a written rule — never as a model inside the product.
01 · Market-regime classification
A classifier labels every stretch of history by its state — trending, ranging, unusually volatile — so a strategy’s results can be read regime by regime, instead of as one average that hides where it fails.
BecomesA written filter: the conditions in which a system trades, and the ones in which it stands aside.
02 · Optimisation & robustness
Search tools explore far more combinations of settings than anyone could test by hand. The single best result is usually a fluke of the sample, so the Lab looks for broad regions where neighbouring settings all work — and tests those.
BecomesDefault settings taken from the stable middle of a region, rather than from its single best point.
03 · News & event analysis
Past economic releases are lined up on their release time and measured: how far price moved, how wide the spread went, how much slippage there was, and for how long. The pattern by type of event matters, not any single headline.
BecomesA measured window around each type of event, written into the specification: when a system stands aside — or, for a news system, when it may act.
Release
Defensive, Moderate and Extreme run the same rules. What changes is how much each one risks — and so how deep a drawdown each is built to accept.
Defensive
Risk
The lowest risk setting: smaller positions, shallower drawdowns.
Moderate
Risk
The middle setting, between growth and drawdown.
Extreme
Risk
The highest risk setting: larger positions, deeper drawdowns. Only for money its owner can afford to lose.
MultiScalper Pro ships all three, and each one runs on its own Myfxbook-verified account. See the three accounts →
After release
The last stage never ends. A released system runs on accounts that Myfxbook verifies — track record and trading privileges — so the record can be checked without taking our word for it.
4Myfxbook-verified accounts on the Performance page
Open the Performance pageCapabilities
MQL4 and MQL5 Expert Advisors built from a written specification, with deterministic risk logic and defensive handling of broker edge cases.
Turning a discretionary approach into explicit rules — including the parts the trader has never had to put into words before.
Tick-level simulation with realistic execution costs, walk-forward harnesses and parameter-sensitivity reports.
Python research pipelines for feature construction, hypothesis testing and statistical validation, ahead of any code.
Economic-calendar data, volatility-regime classification and session-structure analysis — turned into the written rules a system trades by.
Multi-account dashboards, risk supervisors and licensing systems — the operational layer around the strategy itself.
What the Lab will not claim
That any model predicts market direction reliably. It does not, and vendors who claim otherwise are selling something else.
That a backtest is evidence of future returns. It is evidence that a rule set would have behaved a certain way on one historical sample.
That machine learning removes risk. It changes the shape of the risk and adds overfitting as a new way to fail.
That an X-STOX system learns while it trades. The models stay in the Lab; what runs on your account is a fixed set of tested rules.
That any system removes the need to monitor a live account. Automation shifts the work; it does not remove it.
Published research
AI forex bots have evolved fast — and 2025 is the first year where real machine-learning systems finally outperform the old-school “set-and-forget” robots. But the industry is still packed with hype, misleading claims, and bots that look smart on paper but blow accounts in real conditions. This guide cuts through the noise and highlights the top AI forex bots that actually work in today’s market. If you want a bot that adapts to volatility, avoids suicide strategies, and survives news events wit
Discover real user case studies on how AI trading bots are boosting profits in forex trading. Learn about adaptive strategies, risk management, and scalping success with AI-powered bots in 2025.
Discover how machine learning-powered AI bots are changing forex trading in 2025. Learn about predictive analytics, emotional neutrality, and the future of automated forex strategies.
Commission research
Send us the rules. We will put them through the same stress tests and tell you what we find — including if the answer is that it does not survive realistic costs.