Tested: gold strategies, every loss shown
We test popular gold (XAUUSD) strategies and bots against real prices and prop firm rules, and publish the method, the sample size, the drawdown, the costs and the limits. The losing results stay on this page too.
- Rules written firstWhere we can, the rules are saved before any result is seen, and not changed afterwards.
- Costs includedSpread and commission are in every result unless a test says otherwise.
- Unseen dataResults are split into the months a strategy was built on and months it never saw.
- Small samples flaggedFewer than 200 trades means the result is unreliable, and we say so.
Signal Check #01 · tested 2 Nov 2026
The ICT Silver Bullet on gold
Sweep of a liquidity level, displacement, market-structure shift, then entry at the fair value gap, in three one-hour windows a day. Rules pre-registered before the test was run.
No evidence of an edge
| Data | Every real bid/ask tick, 15 Aug 2025 to 11 Sep 2026, built into 1-minute bars |
| Sample | 26 trades (about 2 a month). Too few to trust either way. |
| All months | Win rate 15.4% · profit factor 1.23 · +5.0R in total |
| Months it was built on | 10 trades · +9.3R · profit factor 2.47 |
| Months it never saw | 16 trades · −4.2R · profit factor 0.73 · win rate 6.2% |
| Worst drawdown | 12.3R, all in one losing streak |
| Is it more than luck? | No: p = 0.40 (1.0 after adjusting for the 36 variants checked) |
| Prop firm (FTMO rules) | 0% of challenges passed at normal size |
| Costs | Real spread, $0.07 per oz commission round trip, $0.05 slippage per fill |
| Limits | One market, about 13 months, no news filter, very few trades. |
Correction: an earlier version of our Signal Check #01 video gave different figures (214 setups, 38% win rate). Those figures were wrong. The numbers on this page come from the pre-registered test.
Source: pre-registered protocol and results, ICT test, 2 Nov 2026
Tested #03 · 1 year of real ticks
A +58.8% grid bot year that hid a 44% drop
A gold grid bot with 4 levels: it buys more each time price moves against it and closes the whole basket at a small profit.
Fails prop firm rules
| Data | Every real tick of gold for one year, $10,000 account |
| Result | +$5,882 (+58.8%), profit factor 2.86, 286 trades |
| Balance drawdown | 3.7% |
| Equity drawdown | 44%: the open trades were far deeper underwater than the balance showed |
| Worst single day | 58.5% of the deposit |
| With a daily stop | The same settings lose 17–20% over the year |
| Losing months | 0 of 13, which is exactly why it looks safe |
| Prop firm fit | Fails. A 44% equity drop breaks a 10% max-loss rule many times over, and a 5% daily limit too. |
| Limits | One bot, one market, one year. Did not pass our overfitting check. |
Correction: our Tested #03 video showed "0 of 13" as prop firm challenges passed. It is the number of losing months. The prop firm conclusion is unchanged.
Source: MIDAS backtest review, real-tick replay
Tested #02 · 2 years
FTMO Sweep against prop firm rules
Waits for gold to sweep past a recent high or low, then trades the reversal, with a stop-loss on every trade.
Promising, thin edge, not yet live
| Data | 730 days of gold on 15-minute bars, high-impact news days skipped |
| Account rules | $10,000 model, 5% daily loss, 10% trailing loss, 8% target |
| Result at 0.5% risk | +$6,178, worst drawdown $619 |
| Unseen months | Passed on both the months it was built on and a held-back period |
| Start dates | 31 of 31 passed, but the starts overlap heavily, so this is much weaker evidence than 31 separate tests |
| The catch | At a 2:1 target it lost money (profit factor 0.98). It only turned profitable at 2.5:1 or more, and at 3:1 the spread eats most of the edge. |
| Limits | 15-minute bars, not real ticks. Not yet confirmed on a live account; it is being forward-tested on demo. |
Source: MIDAS research status (730-day funded report) and training findings
Tested #01 · 864 challenges
A gold grid bot through 864 prop firm challenges
72 settings of a gold grid bot, each started in 12 different months, on real ticks.
3 passes in 864
| Data | Every real tick of gold for one year, $10,000 accounts |
| Rules | 10% target, 5% daily loss, 10% max loss |
| Result | 3 passed, 299 failed, 562 still running when the data ended |
| Why | With more risk the grid broke the 10% limit; with less it almost never reached the target |
| Safest setting | Never failed, but passed only once, and that took 116 days |
| Limits | One bot, one market, one year. It does not mean every bot fails. |
Source: prop firm search log, 864 challenge runs
Test your own strategy the same way
Upload an MT5 backtest report or a TradingView list of trades and see how it does against nine prop firms' rules, free.
Open Challenge Replay