Gold Cortex EA is an automated MetaTrader 5 (MT5) robot that trades XAUUSD (gold) on the M15 chart using a Deep LSTM neural network shipped inside the file as an embedded ONNX model. Zerqon EA public MQL5 signal shows 481.4% growth on a $300 account across 1,449 trades, with no money added or taken out. Almost all of that was earned before the robot went on sale, and the file we tested cannot be backtested over the months it has actually been sold. We ran a 100% real-tick backtest and read all 969 trades it placed. Free download at the bottom of this 2026 review.
TL;DR – Free MT5 gold robot with a real, MQL5-verified live account behind it. But 88% of the trades in our test carried no stop loss at all, it stacks up to 5 same-direction positions at worse and worse prices, and its strategy tester replays a recorded sequence that runs out on 29 June 2026.
What is the Gold Cortex EA?
Gold Cortex EA is a free MetaTrader 5 expert advisor that trades XAUUSD on the M15 chart. It carries a trained LSTM neural network as an embedded ONNX model, opens market orders at the top of a 15-minute bar when the model signals, and closes them with a software exit rather than a fixed target. It holds up to 5 positions at once and sits out of the market most of the week.
Core Strategy and Logic
We ran version Zerqon EA 22.002 on a Tickmill demo, MT5 build 6090, over 39,576,157 real ticks, then read the order and deal log rather than the summary page.
- Market orders on the bar open – All 969 entries landed on an exact M15 boundary, 236 at :00, 252 at :15, 241 at :30 and 240 at :45, with nothing in between. Nothing is pending and nothing waits for a break.
- Fixed 0.01 lots – All 1,938 deals were 0.01. The volume never rose after a loss and no position was ever added to a loser, so the no martingale claim is true and the log shows it.
- Software exits, not targets – 846 of the 857 winners were closed by the EA sending a market order. Median holding time was 10.7 minutes and 83.2% of trades closed inside the hour.
- Stacking, in one direction – All 410 trade clusters were single-direction. When a second position opens it is always on the same side as the one already running, up to a maximum of 5.
- Quiet by day, busy by hour – Only 46 trading days produced all 969 trades, so an active day averages 21 of them and most weekdays see none.
The shape is a high-frequency scalper that leans into a move rather than away from it, closer to the pending-order gold scalper in Gold Reflex on trade length and to the Quantum Queen gold grid on how exposure builds.
Please test in a demo account first for at least a week. Also, please familiarize yourself and understand how this Zerqon EA works, then only use it in a real account.
Recommendations for Zerqon EA
- Minimum deposit – Zerqon EA vendor says $600. Read the sizing section below before you accept that, because the lot does not change between $300 and $2,000.
- Pair and timeframe – XAUUSD on M15, which is what both the live account and our test ran. Nothing else has been proven.
- Account type – hedging, low spread ECN or raw. The 50-point spread cap is tight for gold, so a standard account with a wide gold spread will sit idle.
- VPS is required – With no stop loss on most positions, a dropped connection leaves them with nothing holding them (FXVM).
- Work out what a full 5-position stack means on your balance with the lot size optimizer, and check margin per position on the margin calculator (Find the Perfect Broker For You Here).
Performance Review: Live Signal and Our Own Backtest
1. Most of the growth predates the launch
Deposits and withdrawals are both zero, so the headline needs no rebuilding, and 1,449 trades is a bigger sample than almost anything in the gold robot category. Then read the dates. The account’s history starts in February 2026, but the signal was only registered on 8 May 2026 and the robot listed on 25 May. Everything earlier is history MQL5 imported when the signal was created, which is why the growth chart runs the full record while the equity and drawdown panels beneath it start in May.
| Period | What it is | Net profit | Growth |
|---|---|---|---|
| Feb to 31 May 2026 | Imported history, predating signal and launch | +$1,304.10 | +434.7% |
| 1 Jun to 18 Aug 2026 | Monitored and on sale | +$140.10 | +8.7% |
90.3% of that profit was made before the first full month anyone could buy the robot. The green tick has one more gap: the account runs on a server called Leverate-Live, and Leverate supplies platforms to brokers rather than operating as a regulated retail broker. MQL5 confirms the trades happened but does not audit the feed, so the quotes and the fills come from the same side.
2. The backtest is silent for the two months since
We backtested on a Tickmill demo, 100% real-tick, 21 February to 18 August 2026, fixed 0.01 lots on a matching $300. Our copy places no trades at all across the last two months, while the live account traded normally through exactly those weeks. The tester had real ticks to 17 August and 11,458 M15 bars. It made 969 trades to the end of June, then nothing across 36 straight weekdays. Three runs from different start dates all stopped at the identical instant, 2026.06.29 01:45:00, and the log gives the reason away in one word:
Zerqon replay close failed index=4312 ticket=713 retcode=10018 market closedThat index is anchored to the calendar, not the test: 4312 fires at the same timestamp in every run while only the ticket changes. So the EA walks a stored, timestamped sequence, and it empties out at the build date. Its last real activity is 22 and 23 June, version 22.002 is dated 22 June, and two stray trades on 29 June are all that follow. Pre-computing a model’s output is a normal way to make a neural network EA run fast, so we are not calling it a trick. The consequence holds either way: every backtest of this robot ends at the build date of the file you hold. Our 12-point EA scam checklist covers why a tester report is weak evidence.
Past that date, at an unchanged 0.01 lot so dollars compare directly, monthly output falls from $326 across February to May to $16 across July and August. Seven weeks is one bad month and one good one, not a verdict, and 22.105 landed mid-window.
3. What the order log shows
The listing says every trade is protected by a predefined stop loss. In our run 853 of 969 entries (88.0%) carried neither a stop loss nor a take profit. Ninety-four carried a stop only, a consistent $20.97 out, 22 a target only, none both. Software exits only work while the software runs, so if your VPS drops, 88% of open positions have nothing behind them. That belongs to 22.002; the developer describes 22.105 as changing “the updated Stop Loss behavior” while stating “the core entry logic itself was not changed”.
The listing also says it does not use martingale, grid or position averaging. The first is true. The third is not: 558 entries opened while a same-direction position was already running, and 463 (83.0%) opened at a worse price than the average of the open ones. The lot never grows, so risk does not compound like a doubling ladder, but exposure and adverse move arrive together. The worst set opened 6 buys on 11 May for -$97.85.
| Metric | Live signal | Our backtest ($300) |
|---|---|---|
| Closed trades | 1,449 | 969 |
| Profit factor | 1.43 | 1.76 |
| Average win / loss | $3.74 / -$20.66 | $3.99 / -$17.07 |
| Max drawdown | 16.55% balance | 14.74% balance, 17.52% equity |
| Longest losing run | 6 trades, -$126.90 | 4 trades, -$82.98 |
The average loss is 5.5 times the average win, so the live account breaks even at an 84.7% win rate. It is doing 88.75%, so the whole result rests on a 4-point buffer. Check your own deposit against the drawdown calculator before deciding that curve looks calm.
Gold Cortex EA
A free MT5 XAUUSD robot on an ONNX neural network, with a real $300 account that earned 90.3% of its profit before it went on sale.
Pros
- Clean account: $300 deposited, $0 added, $0 withdrawn, so the growth figure needs no rebuilding
- Volume never moved off 0.01 lots across 969 trades, and no losing position was enlarged
- The "up to 5 open positions" and "does not trade continuously" claims both hold
- Profit is spread thin rather than concentrated, with the best 10 of 969 trades producing 11.9% of net
Cons
- 90.3% of the published profit came before the first full month the robot was on sale
- 853 of 969 entries (88.0%) carried no stop loss and no take profit, against a listing that promises one on every trade
- 463 of 558 stacked entries opened at a worse price than those already open, which is the position averaging the listing denies
- Average loss is 5.5 times average win, so it needs an 84.7% win rate to break even
- News filter and both prop firm guards ship off, and the signal sits on a Leverate-Live server rather than a named regulated broker
Summary
The account is clean: $300 in, nothing added, 1,449 trades, now $1,744.20. The dates are the problem. The signal was registered 8 May 2026 and the robot listed 25 May, so $1,304.10 of the $1,444.20 predates the first full month anyone could buy it. Our run also found no stop loss on 853 of 969 entries.
What we run on our own charts
Every EA in this review is someone else’s code. FXC Fusion is ours: an MT4 and MT5 breakout indicator built around one graded trade a day, scored Q1 to Q5, with the entry, stop loss and three take-profit levels drawn on the chart before you commit. No basket, no recovery ladder, no strategies competing for the same level.
Here is one full session from the account we run it on, 12 August 2026 across 14 pairs.
One session: 41 trades closed on 12 August 2026 across 14 pairs. Pips are as traded; dollar figures are scaled to a uniform 0.10 lot so every pair is comparable. Open positions excluded, net of commission and swap. Gold and crypto pips normalised to $0.10 and $1.00 per pip.
See FXC Fusion →There is a monthly plan from $79 if you want it on your own charts before paying for a licence, and your most recent payment comes off the price when you switch to lifetime. See a real trading day with it or view FXC Fusion.
Frequently Asked Questions (FAQ)
Leverate-Live server, and Leverate supplies platforms to brokers rather than being a regulated retail broker, so quotes and fills come from one side.Conclusion
Gold Cortex EA has what most free gold robots do not: a real $300 account, no top-ups, 1,449 trades and a profit that needs no rebuilding. However, $1,304.10 of it came before the first full month it was on sale, the signal sits on a platform supplier’s server, our run found no stop loss on 88% of entries, and the backtest goes silent for two months while the live account kept trading. What is left to judge is 8.7% since launch on a system that breaks even at an 84.7% win rate.
