Gold Snap is structurally one of the cleanest EAs in our catalog. It holds at most one position at a time. Stacking entries 0. Entries added against a losing position 0. Moments with a long and a short open together 0. Volume increase after a loss 0.0%, after a win 0.0%. Lots are fixed at 0.01. Four of our six verdict dimensions come out ok, including prop fit, which most of this catalog does not manage.
It also wins 61.9% of its trades, and in the most recent years far more than that. 2025 closed 91.3% of its trades in profit. And the entire 23 year run made +$1,922.00.
The run contains 3,769 trades, which is roughly 160 trades a year. Divide the result by the count and you get $0.51 per trade. Our sensitivity section reaches the same number independently. The break even shock, the extra cost per trade that would flatten the whole audit, is $0.51.
The cost ladder is where this becomes concrete. Add $0.50 of cost per trade and the 23 year result falls from $1,922.00 to +$37.50 at a profit factor of 1.01. Add $1.00 and it is -$1,847.00. Add $2.00 and it is -$5,616.00. The measured average position is 0.01 lots, so half a dollar per trade is not a stress scenario. It is a slightly wider spread than the one in our test data.
The yearly table has a direction. 2003 loses $86.08 at a win rate of 17.0%. 2004 loses $141.13 at 18.5%. 2005 loses $160.22 at 18.9%. 2006 loses $74.29 and 2007 loses $50.13, both around 47%. From 2008 onward the win rate never falls below 64% again, and it ends at 91.3% in 2025 and 84.3% in the partial 2026. Only two later years are negative, 2017 at -$0.71 and 2018 at -$6.67, which are rounding errors rather than losses. That is 7 negative years in total and it earns a caution in our regime dimension.
A win rate that goes from 17% in the first three years to above 80% in each of the last four is not how a stable rule set behaves on a stable market. It is how a rule set behaves when it fits the market it was built on and does not fit the market of 20 years ago. We cannot prove tuning from a report and we are not claiming it. What we can say is that the two most recent years, 2025 and 2026, contribute $779.56 of the $1,922.00 total, which is 40.6% of everything, from 197 trades out of 3,769, which is 5.2% of them.
Our measurement frame is a 100,000 USD account, because that is what makes a 23 year run readable. The vendor recommends 200 USD, and our engine takes that seriously enough to replay the same deal sequence against it. On a 200 USD deposit, 14.6% of the 7,538 possible start points end in a dead account. In 2003 the figure is 100%. Every single start point in that year runs out of money.
The chained version of the same test is even clearer. Starting on 2003-05-08 with 200 USD, the account dies on 2004-11-19 after 560.9 days. Restarted, it dies again on 2006-02-24 after 462.0 days. The third attempt survives the remaining 7,474 days to the end of the run. Same EA, same settings, same deposit. The only difference is the starting date.
None of that shows up in the prop fit dimension, and correctly so. Against the IQ Capital Classic rule set our sweep finds 0 account deaths at every size from full down to one eighth, 0 breaches of the per position loss limit and no consistency breach in any year. On a funded 100,000 USD account this EA is quiet enough to be invisible. On the account the vendor suggests, it is a different instrument.
Our withdrawal replay sweeps everything above the start balance out on the last trading day of each month. Across 280 months it paid in 109, with a median paid month of $13.78 and a best month of $91.31. The longest run without a payout is 115 months. That is nine and a half years, and it sits at the front of the backtest, where the losing era and the two account deaths are.
The profit is also concentrated. 80% of it was made in 169 days out of 2,603 trading days with trades, which is 6.5% of them. The best single day made $24.01 and the worst lost $20.93, so nothing here is a jackpot. It is a thin edge that only shows up on a small number of days.
The same vendor also sells Gold House, which we audited on the same symbol, the same period and the same measurement frame. The shape repeats. Gold House loses in 2003, 2004, 2005 and 2006 and is positive in every year afterwards. Gold Snap loses in those four years plus 2007 and is positive in almost every year afterwards. Both are XAUUSD systems that fail on the gold market of 2003 and work on the gold market of 2025.
Where they differ is size of edge. Gold House earns $1.06 per trade across 10,402 trades. Gold Snap earns $0.51 across 3,769. Both numbers are small, and one of them is half the other. If you are choosing between the two, that ratio matters more than any screenshot on either sales page.
Our engine tracks what the instrument itself did over the same window. Gold went from 355.33 to 4,278.64 in the entry price series of this run, a move of 1,104.12%. Buying 0.01 lots at the start and holding to the end would have made $3,923.31. The EA made $1,922.00, which is 49% of that. The EA did it with a deepest reconstructed drawdown of 0.54%, and holding gold for 23 years means sitting through every gold crash of that period, so the risk adjusted comparison favours the EA. The plain return comparison does not.
The run itself is clean. 100% real ticks from 2003.05.05, 1,409 absent minutes out of 7,899,407 minute bars, 0 failed entries, 1 failed modify attempt and 0 randomizer prints, so the run is deterministic and repeatable. Log coverage is 1.0, chronology violations 0, duplicate deal ids 0, FIFO deviation from the journal 0, and the reconstructed final balance matches to the cent at $101,922.00. Commission across the run was -$301.52 retrofitted at $3.50 per lot per side and swap was -$9.17 across 33 overnight trades, together 13.9% of gross profit.
Note that last figure against the 8.1% we measured on the sibling EA. When the edge per trade shrinks, the same commission eats a larger share of it. That is the whole risk of this product in one sentence.