Smart Gold Impulse wins almost nine trades out of ten and keeps 22 cents from each one. Our run closes 4,163 trades on gold, 88.3% of the buy setups and 88.4% of the sell setups end in profit, and the account finishes 926.20 USD above where it started. Divide the one number by the other and the whole system fits into a single figure. The average trade is worth 0.22 USD. Our cost sensitivity says the same thing from the other side. An extra 0.22 USD of cost per trade takes the result to zero, because that is the same arithmetic written backwards.
One deterministic tester run of version 1.00 on XAUUSD# M15, 100% real ticks, first trade 2006-04-20 and last trade 2026-08-12. We reconstruct the run deal by deal against the tester journal, so pairing confidence is exact, journal coverage is 1.0, chronology violations 0 and duplicate deal ids 0. The reconstructed balance matches the reported one to the cent. The catalog protocol uses a fixed 0.01 lot on a 100,000 USD frame at 1:5000, which is a measurement setting and not a recommendation. On top of the tester result we retrofit commission at 3.50 USD per lot per side, which costs this run 333.04 USD across 8,326 deals. The vendor header before that retrofit reads 1,259.24 USD at a profit factor of 1.22.
Gross profit before costs is 1,257.61 USD. Commission takes 333.04 USD and swap adds 1.63 USD, so 26.4% of the gross never reaches the account. Our audit rates the cost dimension caution for exactly this reason. A system that hands over a quarter of its gross to friction has no room left when the friction changes.
The sensitivity table puts a number on that. Half a dollar more cost per trade turns +926.20 USD into -1,155.30 USD at a profit factor of 0.81. One dollar more takes it to -3,236.80 USD. For comparison, the commission we retrofit costs this run 0.08 USD per trade, so the half dollar shock is more than six times the friction we already charge, which is why we retrofit commission at all rather than trusting a commission free tester setting.
The tester window on the report runs from 2003.05.05, and this is where a backtest chart and a trade census disagree. The first trade lands on 2006-04-20, almost three years into the offered window. 2007 carries no trade at all. 2017 carries exactly one, and it loses 12.17 USD. Then the pattern reverses. 2025 produced 1,163 trades and 2026 produced 1,967 trades by 12 August. Together that is 3,130 of the 4,163 trades, or 75.2% of the whole record, inside the final two calendar years of the run.
The profit follows a different curve than the trade count, which is worth sitting with. The nineteen years from 2006 to 2024 hold 1,033 trades and 540.03 USD. The year 2025, the busiest full year in the record with its 1,163 trades, contributed 3.96 USD. The partial year 2026 contributed 382.21 USD, which is 41.3% of everything the system made in twenty years.
Density tells the same story from another angle. The account has a closed trade on 904 days. Between the first and the last trade there are 5,299 weekdays in our end of day series. The robot is flat on roughly five days out of six. We do not know why the frequency jumped, because a tester report does not explain a decision rule. What we can note is that the input panel works in absolute points. Base stop 1,200 points, trailing start 150, trailing distance 50, minimum distance between orders 150. Over the traded span gold moved from 765.83 to 4,407.67, a rise of 475.54%, so the same point distance meant something very different in 2006 than it does now. That is an observation about the settings and the market, not a claim about the code.
The exit census answers it in one line. Of the 4,163 trades, 4,146 ended at a stop and 17 closed by signal or time. Not one closed at a take profit. Of those stop exits, 3,672 closed in profit, which is 88.6% of them. That is a trailing stop harvesting small gains, and the high win rate is a property of that mechanism rather than evidence of prediction quality.
The structure is otherwise clean and our audit says so. No martingale, no adverse averaging, 0 adds against the trend, never more than 2 positions open at once and 0 moments with a long and a short at the same time. Buy and sell setups split almost evenly at 2,101 and 2,062 trades and both make money, 340.93 USD and 585.27 USD, which add up exactly to the 926.20 USD net. This is not a hidden risk product. It is a thin one.
80% of the profit was made on 29 days, which is 3.2% of the 904 active days. The best single day earned 36.68 USD on 2026-05-06. The worst single day lost 69.26 USD on 2025-05-01, so the worst day is close to twice the best day. Our end of day analysis counts 42 drawdown episodes. The deepest one peaked on 2026-03-23, bottomed on 2026-04-28 at 252.31 USD below the peak and recovered on 2026-06-24, which is 67 trading days underwater and 0.25% of the measurement frame.
Two ratios frame the whole run. The report header prints a Sharpe of 8.29, measured per trade. Our end of day Sharpe on the same run is 0.63. Compound growth on the 100,000 USD measurement frame is 0.04% per year. None of these are contradictions. They are the same result seen at different resolutions, and the coarser the resolution, the smaller it looks.
The vendor names no recommended deposit, so our capital replay falls back to 1,000 USD. On that basis none of the 8,326 possible start points ends in a dead account, which is a genuine point in the product's favour. The same replay shows the other side. Re-expressed against 1,000 USD the observed path carries a maximum drawdown of 25.23%, and all three prop firm floors we model are breached on the observed path itself, not merely in simulation.
On the prop firm side our replay flags the consistency rule. Best day share exceeds the allowed percentage in 14 years under the funded profile we model, which is structural rather than unlucky. The prop fit lamp is red.
Over the traded span the entry price series of this run rises from 765.83 to 4,407.67. Holding 0.01 lots through those twenty years would have earned 3,641.84 USD. The EA earned 926.20 USD on the same lot size, so roughly a quarter as much, spread over 4,163 trades. That comparison is approximated from entry prices and holding gold means sitting through every gold crash since 2006, so it is not a fair risk comparison. It is a scale comparison, and scale is the honest question when a system asks for 4,163 trades of your attention.
Smart Gold Impulse was published on 2026-07-04 and has not been updated since. Our test window ends on 2026-08-12, so there is a real out of sample stretch, and it is short. 39 days after release hold 268 trades and +146.46 USD at a profit factor of 1.52 and a 90.7% win rate, which is better than the 1.14 profit factor of the 3,895 trades before release. We report it and we do not weight it. 268 trades over 39 days decide nothing about a system whose record spans twenty years, and the out of sample window here is shorter than a single quiet quarter.
One date does count in the product's favour. We measured version 1.00 and the market listing still shows version 1.0, so this audit describes the software that is actually on sale. That is not always the case, which is why every audit we publish names its version and its date.
The deal level census, the concentration analysis and the capital replay need full deal reconstruction, which is what we sell. Any catalog audit is $19, and EAs with several audited presets or versions come as one bundle. Your own tester report? The browser check is free.