Cupella Gold Triad MT5
- 专家
-
Oguzhan Senturk
Systems Architect & Algorithmic Trader
Engineering fully autonomous execution systems, institutional risk engines, and bespoke trading software.
ABOUT ME & PHILOSOPHY - 版本: 1.0
- 激活: 10
CUPELLA GOLD TRIAD runs three separate trading engines on a single XAUUSD account. Each engine has its own timeframe and its own working conditions, so they rarely open positions at the same time. Running all three on one balance is a deliberate choice: it is why the account curve is steadier than any single engine on its own.
In our nine-year MetaTrader 5 Strategy Tester simulation on real tick data, a 10,000 USD account reached 126,198 USD with a maximum drawdown of 18.65%. Commission and spread were charged on every trade. Every figure on this page comes from that simulation, and we show the weak years alongside the strong ones.
MEASURED RECORD
Nine years, run with every tick based on real ticks modelling, the highest-accuracy model the platform offers. We took the spread from the historical tick stream rather than assuming a fixed value, and charged 7 USD round-turn commission per lot on every position.
| Measurement | Result |
| Test period | January 2018 – September 2026 (8.7 years) |
| Account growth | 10,000 → 126,198 USD |
| Compound annual return | 34.0% |
| Maximum drawdown | 18.65% |
| Average monthly return over nine years | 2.70% |
| Calmar ratio (return over drawdown) | 1.82 |
| Total trades | 2,300 – about 22 per month |
| Trading engines | 3, running on one account |
| Instrument | XAUUSD only |
To be clear about one thing: 2.70% is a nine-year average, not a monthly target. Months and years differ considerably, and the year-by-year table below shows all of them.
THREE ENGINES, ONE ACCOUNT
Most gold robots run a single idea. When that idea stops matching the market, the account stops working. That is why we spread the work across three engines. We developed and measured each one separately, and only combined them after seeing that their results were largely independent of each other.
| Engine | Holding period | Contribution to the portfolio |
| Engine 1 | Several days | The slowest of the three. Fewest positions, longest holding time, largest average result per trade. |
| Engine 2 | Within the day | Opens and closes inside the same trading day. Carries no overnight financing cost. |
| Engine 3 | Hours | The fastest of the three. Highest number of positions, shortest exposure per position. |
The three operate on different clocks, which is what keeps them from committing the account to the same idea at the same time. The specific conditions each engine trades are part of the product and are not published.
Why this matters: we also ran each engine on its own over the same nine years, and measured how their monthly results relate to each other. The correlations came out at +0.004, -0.001 and +0.222 - close to zero. A flat period for one engine is usually not a flat period for the others.
| 10,000 USD over nine years | Result | Deepest drawdown |
| Engine 1 on its own | 35,260 USD | 15.59% |
| Engine 2 on its own | 13,820 USD | 5.99% |
| Engine 3 on its own | 23,198 USD | 16.75% |
| All three together | 126,198 USD | 18.65% |
This is the core of the product. The best single engine reaches 35,260 USD on its own. Together they reach 126,198 USD, while the deepest drawdown rises only from around 16% to 18.65%. The return multiplies; the risk does not.
Each engine sizes its positions from the distance to its own stop, so a wider stop means a smaller position, not a bigger loss. Increasing position size after a loss is not something this EA does.
HOW IT WAS TESTED
We did not build this by turning dials until the backtest looked good. It was developed in a separate research environment and then confirmed in the Strategy Tester. Here is what we applied along the way:
| Test | Purpose |
| Out-of-sample testing | Rules were fixed on one period and then measured on periods that were not used to build them. |
| Shuffled controls | The same rules were run against randomised data to confirm the result was not produced by chance. |
| Block-shuffle controls | A second control that preserves the trend structure of the data, to separate a genuine effect from a simple directional bias. |
| Deflated Sharpe ratio | The performance statistic was adjusted for the number of variations tested, so that a result reached after many attempts is not treated as if it were the first. |
| Delay tests | Every signal was re-measured with an execution delay to confirm that no result depended on acting on information before it was available. |
| A/B determinism runs | Identical configurations were run twice and compared to the cent. Every change to the product was verified against an unchanged baseline the same way. |
| Cost sensitivity sweeps | The full period was re-run at rising spread levels to establish the trading conditions under which the result holds. |
| Benchmark comparison | Results were compared against buy-and-hold gold over the same period with financing costs included. |
| Correlation testing | Each engine was also run on its own over the same period and the correlation of monthly results measured: +0.004, -0.001 and +0.222. That is why they ship together. |
| Market regime breakdown | Results were separated into rising, falling and sideways gold markets and required to hold in each, rather than depending on a single market condition. |
| Financing cost audit | Overnight swap was measured separately from trading profit, because a strategy can look profitable on paper while financing consumes the result. |
| Position sizing floor test | Account sizes were tested downward until the broker minimum lot size began to distort the intended risk. This is where the 2,500 USD minimum comes from. |
| Random start-date testing | The strategy was started at many different points in the period, so the result does not depend on one favourable starting date. |
| Shipping build verification | The published figures were re-measured on the exact compiled file sold here, not on a development version. |
Ideas that failed these controls were dropped rather than adjusted until they passed. The three engines we kept are the ones that held up.
YEAR BY YEAR — INCLUDING WEAKER YEARS
| Year | Return | Note |
| 2018 | -8.9% | Only negative year |
| 2019 | +61.4% | |
| 2020 | +82.1% | Strongest year |
| 2021 | +2.2% | Quiet year for gold |
| 2022 | +54.3% | |
| 2023 | +8.9% | |
| 2024 | +53.2% | |
| 2025 | +55.1% | |
| 2026 (through September) | +15.4% | Partial year |
Eight of the nine years closed positive. 2018 closed at -8.9%, the only negative year in the record. 2021 and 2023 were quiet years rather than losing ones.
We show 2018 and 2021 deliberately. A nine-year record with no weak years would not be a realistic one. What the table shows is that the weak years here were shallow: the strategy went quiet rather than losing badly, and only one of the nine years closed negative.
COMPARED WITH SIMPLY HOLDING GOLD
Gold rose a long way over this period, so the fair question is what an EA adds over just buying gold and holding it. Both columns below cover the same instrument and the same dates.
| January 2018 to September 2026 | Holding gold | CUPELLA |
| 10,000 USD becomes | 33,218 USD | 126,198 USD |
| Total return | 232% | 1,166% |
| Compound annual return | 14.86% | 34.0% |
| Maximum drawdown | 26.60% | 18.65% |
| Return per unit of drawdown | 0.56 | 1.82 |
| Negative years | 3 of 9 | 1 of 9 |
The result is 3.8 times larger, and the drawdown is shallower. The point worth noting is that the gap was not closed by taking more risk. Holding gold returned less and fell further.
Worth adding: gold peaked in January 2026, fell 26.6% into July 2026, and had not recovered that peak by the end of the test period.
| Difference | Holding gold | CUPELLA |
| Direction | Profits only when gold rises | Takes positions in both directions |
| Loss limit | None. The position is held through any decline. | Every position is opened with a stop. |
| Position size | Fixed by the amount purchased | Calculated per trade from the account and the stop distance |
| Flat markets | Capital stays tied up and returns nothing | 2021 and 2022 returned close to zero for gold; the EA returned +2.2% and +54.3% |
| Financing | A leveraged gold position pays overnight swap continuously | Two of the three engines close within the day, so most positions carry no overnight cost |
Look at 2022 in particular: gold ended that year at -0.35%, CUPELLA returned +54.3% over the same twelve months.
RISK CONTROL
Position size comes from account equity and the distance to the stop, so the money at risk stays proportional to your account rather than to the market. No grid, no martingale, no averaging down. Positions are not enlarged to recover a loss.
| Risk profile | Intended use |
| Conservative | Lower position sizing throughout. Shallower drawdown and a slower curve. |
| Balanced — default | Every figure on this page was measured at this setting. |
| Dynamic | Slightly higher sizing than Balanced, capped so that the measured drawdown ceiling is preserved. |
You select the profile on the panel and can change it while the EA is running. The account is never exposed to a higher risk level than the profile you chose.
WHICH BROKER TO USE
Use a broker whose average XAUUSD spread is 0.15 USD per ounce or lower. This means a RAW, ECN or Zero-spread account type.
| Broker | Account type |
| IC Markets | Raw Spread |
| Pepperstone | Razor |
| Tickmill | Raw |
| FP Markets | Raw |
Pricing changes over time and this is not an endorsement of any broker. If you use a different broker, send a message before purchasing and it can be checked.
REQUIREMENTS
| Item | Requirement |
| Symbol | XAUUSD only. The EA will not start on another instrument. |
| Chart timeframe | Any. The EA reads the timeframes it needs internally. |
| Account type | RAW, ECN or Zero-spread |
| Minimum balance | 2,500 USD. 5,000 USD or more is recommended. |
| Account currency | USD recommended |
| Leverage | 1:100 or higher |
| Platform | MetaTrader 5, hedging account |
| VPS | Recommended for continuous operation |
About the minimum balance: 2,500 USD is a floor we measured, not a marketing figure. Below it, the broker minimum lot size stays larger than the position the risk calculation asks for, which pushes the real risk per trade above the level we intend. That is why we do not quote a lower number.
CONTROL PANEL
A chart panel is included, available in English or Turkish.
- Equity curve, monthly and annual return
- Current drawdown and maximum drawdown to date
- Open positions and recently closed trades
- Risk profile selection
- Built-in guide pages and help buttons
We added a single-instance lock so a second copy of the EA cannot duplicate orders on the same account. The broker server time offset is measured from the server rather than assumed.
SETUP
Attach CUPELLA to one XAUUSD chart, pick a risk profile and enable Algo Trading. One chart is enough; the EA manages all three engines from it. Do not run a second copy on the same account.
There are five settings in total. The strategy itself has no adjustable parameters: the configuration you receive is the configuration that produced the record above, and it cannot be knocked out of tune by accident.
| Input name | What it does |
| Risk level | Three measured steps: Conservative, Balanced (default), Dynamic. All three stay under the 20% drawdown ceiling. |
| Allocated capital | Trade only part of the account. 0 (default) uses the whole balance. Enter an amount to size everything from that figure instead. |
| Language | Panel language, English or Turkish. |
| Panel corner | Which corner the panel sits in. |
| Show panel | Panel on or off. |
RUNNING IT ON PART OF YOUR ACCOUNT
You may not want to commit a whole account to one system. Open the Inputs tab, find the setting named Allocated capital, and type the amount you want to commit. If you hold 50,000 USD and want CUPELLA working with 10,000 of it, enter 10000 there. Leave it at 0 and the whole balance is used, which is the default.
Position sizing, the exposure limits and every figure on the panel then work from that amount, so the risk you agreed to is the risk you get — there is no need to open a separate account. The panel shows both numbers side by side: your account balance, and the capital allocated to CUPELLA underneath it.
The allocation is not a fixed number that stays put. It moves with what CUPELLA itself earns or loses, so a losing run reduces position size exactly as it would on a dedicated account. Your deposits, your withdrawals and any other system you run are excluded from the calculation.
Three things worth knowing. The allocation cannot exceed your real balance; if you enter more than you have, the EA uses what is actually there. The 2,500 USD floor applies to the allocation rather than to the account, so allocating less than that runs into the same minimum-lot problem described above. And because the panel's return and drawdown figures are percentages of the allocated capital, changing the allocation starts a fresh tracking period — the panel says so when it happens, and your trade history is untouched. Change it when nothing is open, not in the middle of a position.
PRICING
The price is tiered. As the number of licenses sold increases, so does the price, so buying earlier costs less.
| Tier | Licenses | Price |
| FOUNDER — current tier | 1 – 10 | 299 USD |
| Second tier | 11 – 25 | 349 USD |
| Third tier | 26 – 50 | 599 USD |
| Final tier | 51 and above | 899 USD |
A purchased license includes future updates at no additional charge. There is no recurring fee.
RENTAL
If you would rather try it before buying, a rental option is available: 69 USD for one month or 129 USD for three months.
The rental is a short-term option. It will be removed once either 20 rentals or 10 permanent licenses have been sold, whichever comes first. After that the product is available by purchase only.
TEST IT YOURSELF
You can download the demo and run it in the Strategy Tester on your own broker's XAUUSD data.
- Run different date ranges. 2018, 2020, 2022, the last twelve months — the crash months, the flat months, the years gold did nothing.
- Turn on visual mode. You see every entry, every stop, and the panel updating trade by trade.
SUPPORT
If you have a question or want help with setup, send a message through MQL5.
RISK WARNING. All performance figures on this page come from historical simulations in the MetaTrader 5 Strategy Tester and are not live trading results. Past performance is not indicative of future results. Trading leveraged instruments involves substantial risk and you can lose money. Do not trade with capital you cannot afford to lose. Nothing on this page is a guarantee of profit.
