Trades analysed
live0
Every closed trade, with its full path from entry to exit.
Phase 1 of 3 Simulated capital, live markets
An AI trading system for portfolio management, day trading and investing.
QuantX is an open research lab. Our system trades live crypto, forex, commodity and index markets with simulated capital, 24 hours a day, with no human in the loop. It predicts, sizes and manages every position by itself, and every decision it makes is published the moment it happens.
Not for sale. Not investment advice. Every trade public.
Capital under research€20,000
Open positions0
Trades analysed0
The full dashboard is a live application. On a phone we do not load it in the background, so your data allowance stays yours. Open it when you want it.
Net result
0.00%€0
Against every euro paid into a desk since launch. From here on, losses never reset.
Fees paid
0% of capital€0
The line most backtests quietly leave out. Ours lost to it first.
Running
autonomous0h
No human in the loop. No pause button. It has not stopped since launch.
What we are building
One system, three jobs.
Most trading tools do one thing. We are researching whether a single learning system can hold a portfolio, trade it intraday, and still make decisions that make sense over years.
Portfolio management
Capital is split across four independent desks and allocated between six strategies by Thompson sampling. The system moves money towards what is working and away from what is not, without anyone telling it to.
Day trading
On every closed one-minute candle it recomputes indicators, learns the sample that just matured, manages open positions, and only then looks for new entries. Stops and targets are checked on every single tick.
Long-term investing
Winners are not closed at the first target. Stops follow the move, partial profits bank the risk, and targets extend while the model still sees room. That is what lets a good position become a good investment.
How the system works
Six mechanisms, running every minute.
Nothing here is a black box we cannot describe. This is the full loop the engine runs, continuously, on every instrument of every desk.
Prediction engine
An online neural network reads 23 features per instrument per minute: returns over 1, 3, 5, 15 and 60 minutes in volatility units, RSI, EMA distances, MACD, Bollinger and VWAP z-scores, volatility regime, volume shock, real taker buy pressure, move efficiency, Donchian position, time of day, cost and funding. It predicts up, flat or down over a five-minute horizon, with news sentiment and the economic calendar feeding in alongside. Crucially it predicts before it learns each sample, so every hit rate we report is genuinely out of sample.
Self-improvement
The model learns from every matured sample while the market is open, and the capital allocator learns from every closed trade. A strategy that stops working loses weight within hours. The model only earns the right to influence trades after 600 samples, 150 directional calls, a hit rate above 53 % and a positive edge after costs. Until then it runs in shadow mode, predicting and being scored without trading.
Active trade management
An entry is the beginning, not the decision. The stop moves to break-even plus costs once the trade is far enough ahead. Partial profits are taken at defined R multiples. An ATR trailing stop follows the move. A time stop closes anything going nowhere. And when the model still sees upside, the take-profit is pushed out by a further R while half the open gain is locked in behind it.
Cost and margin realism
A simulation that ignores costs will tell you anything you want to hear. Orders fill at the live bid and ask and pay the spread. Size beyond top of book pays market impact. Every fill is charged taker or maker fees. Perpetuals settle real Binance funding every eight hours; CFDs pay overnight financing at rollover. Margin is modelled as an exchange would: cross margin per desk, leverage caps, an initial-margin check before every order, and liquidation at maintenance margin.
Risk control
Each order passes the same gate: market genuinely open, no news lock, daily loss limit intact, emergency stop clear, position and per-symbol caps respected, correlated-exposure cap respected, cooldown elapsed, quote fresh by exchange timestamp, spread not blown out, target worth several times the round-trip cost, and margin actually available. Size is risk percent of equity scaled by conviction, strategy weight and a drawdown brake, capped at 40 % of free margin.
Continuous backtesting
The same desk code is replayed against real history on a schedule: crypto over 7 days every six hours and over 30 days nightly, the CFD desks over 14 days each morning. Training and test split 70/30, starting from the live model. Results are published whether they flatter us or not.
Findings so far
We publish the losses too.
A research lab that only shows its good runs is marketing, not research. These are the real results that shaped the system you can watch right now.
First 7-day crypto backtest: −51 %
514 trades, 34 % hit rate. The decisive detail: gross profit before fees was roughly zero, and the €1,650 of fees was the entire loss. We responded by gating the AI behind a proven net edge, demanding targets of 3.5 to 5 times cost, and capping correlated positions.
Second 7-day backtest after the tightening: −50 %
381 trades, profit factor 0.40, €1,482 in fees. The model correctly stayed in shadow mode at −11.9 bps net edge per call. The honest read: the rule-based strategies have no gross edge on the 1-minute and 5-minute timeframes, about −€0.89 per trade before fees, against €3.89 of fees per trade.
One-minute scalping does not survive taker fees
At roughly 10 bps round trip there is no room for an edge this small. The research moved on: maker entries, 15-minute and 1-hour timeframes, funding and basis strategies, regime filters.
Where the second backtest actually went
7 days of crypto, 381 trades. The strategies were close to flat before costs. The fees were the entire loss.
Gross P&L
about €0
Fees paid
−€1,482
Net result
−50 %
| Friction | How it is modelled |
|---|---|
| Spread | Every market order fills at the live bid or ask |
| Market impact | Applied when size exceeds top-of-book depth |
| Crypto fees | Binance USD-M VIP 0: taker 0.050 %, maker 0.020 % |
| CFD commission | NAGA: 2.50 € per side on single-stock CFDs, none on indices or FX |
| Funding | Real Binance perpetual funding, every 8 hours |
| Financing | Benchmark + 3.5 % p.a. (ECB 2.65 % → 6.15 %), booked daily |
| Margin | Cross margin per desk, initial-margin check per order |
| Leverage | Capped per instrument, rejected above the cap |
| Liquidation | Everything closed when equity ≤ maintenance margin |
| Trading hours | CFDs Sun 22:00 to Fri 21:00 UTC, US stocks 13:35 to 19:55 UTC |
| Friday flat | CFD positions closed before the weekend gap |
Every item in this table is enforced in the engine, not an aspiration. Watch it apply in real time in the activity log.
Where these numbers come from
We do not estimate costs, we copy them from the venues we actually use. The crypto desk is priced at Binance USD-M VIP 0. The forex, commodity and index desks are priced at NAGA, the broker this research is built on: no commission on FX, indices or metals, 2.50 € per side on single-stock CFDs, and overnight financing at the benchmark rate plus 3.5 % per year.
Why it matters: in our first live sample the gross result before costs was +0.69 € while fees came to 206.44 €. If the cost model is wrong by even a little, every conclusion drawn from it is worthless.
The NAGA link above is a referral link. It costs you nothing and pays us a commission if you open an account. We mention it because it is our own broker, not because anyone paid for placement, and it has no bearing on the results published here.
Roadmap
Three phases. We are in the first one.
We will not put real money behind this system until the evidence says we should. That is the whole point of doing it in public.
Phase 1 · Running since 11 October 2026
Simulation on live markets
Live prices, simulated capital, full cost and margin realism, no human intervention.
- Four desks, €5,000 each
- Every trade and adjustment published
- Continuous automated backtests
- Model gated behind a proven net edge
Phase 2 · Next
Live portfolio, real capital
Only once the AI delivers consistent, statistically meaningful positive results under risk management.
- Hard risk limits from day one
- Small size, scaled by evidence
- Same public reporting as today
- Exchange API, real execution
Phase 3 · Later
Opening the research up
If, and only if, the out-of-sample record holds up over a long period.
- Signal and strategy API
- Access for selected partners
- Investor and press briefings
- Published methodology
Register your interest
Be there when QuantX opens up.
QuantX is not a product you can buy today, and we are not taking money from anyone. What we are doing is building a record in public. If you want to know the moment that record turns into something you can use, or if you want to talk to us before then, leave your address and tell us which of these you are.
- Follow the research. A short, honest update when something meaningful changes, not a weekly newsletter for its own sake.
- Investors. We will share the full out-of-sample record and methodology when Phase 2 opens.
- Press. Happy to talk about what an AI trading system actually does, including the parts that lose money.
- Partners. Data providers, brokers and exchanges: we are interested.
Questions
The things people actually ask.
Is QuantX trading real money?
No. QuantX is in Phase 1: the system trades live market data with simulated capital. Real-money trading is hard-disabled in the code, and it will stay that way until the AI shows a consistent, statistically meaningful net edge under full risk management.
Can I buy QuantX, or invest in it?
Not today. QuantX is a research project. It is not for sale, we are not raising, and nothing on this site is investment advice or an offer of any kind.
If you want to be told when that changes, whether that is access, a signal API, or an investor conversation, the waiting list above is the way.
Does the simulation actually account for trading costs?
Yes, and that is most of the point. Orders fill at the live bid and ask and pay the spread, size beyond top of book pays market impact, and every fill is charged taker or maker fees. Perpetuals settle real Binance funding every eight hours; CFDs pay overnight financing at rollover.
Margin is modelled the way an exchange does it: cross margin per desk, a leverage cap per instrument, an initial-margin check before every order, and maintenance margin that liquidates the account if equity falls through it. Our first backtests lost money almost entirely to fees, so we are not about to model them away.
How does the prediction engine work?
An online neural network reads 23 features per instrument per minute and predicts up, flat or down over a five-minute horizon. A move only counts as up or down if it exceeds 1.2 times the round-trip cost. Predicting noise that you cannot trade profitably is worthless.
It predicts before it learns each sample, so every hit rate we report is genuinely out of sample. It only earns the right to influence trades after 600 samples, 150 directional calls, a hit rate above 53 % and a positive edge after costs.
What happens to a position after it is opened?
It is actively managed. The stop moves to break-even plus costs, partial profits are taken at defined R multiples, an ATR trailing stop follows the move, and a time stop closes trades that go nowhere. When the model still sees upside, the take-profit is extended by a further R while half of the open gain is locked in behind it.
Every one of those adjustments appears in the public activity log at the moment it happens.
Why run it in public at all?
Because a track record nobody could verify is worth nothing. Running it in the open means we cannot quietly restart after a bad week, and it forces the honest accounting: net profit is measured against every euro paid into a desk, so losses do not disappear through a reset.
The published record starts on 11 October 2026, the day the desks went live. What ran before that was a shakedown of the engine against real prices, not a track record, and we do not count it as one. Everything from the launch date onward stands, good or bad.
Who is behind QuantX?
QuantX is a product of TTA Technologies, a technology company based in Luxembourg.
Important
QuantX is a research project operating on simulated capital. Nothing on this website is investment advice, a recommendation, a solicitation, or an offer to buy or sell any financial instrument. The system is not available for purchase and no investment is being offered or accepted.
Past or simulated results are not indicative of future results. Simulated performance has inherent limitations and does not represent actual trading. Trading leveraged instruments carries a high risk of loss.