This backtest needs no argument about whether it is optimistic. It is negative. Our engine dissected a strategy tester run of Scalping Robot Pro v4.0 over the full available gold tick history and found 302,798 trades that finish below the starting capital. The tester head prints a profit factor of 0.88 and an expected payoff of -0.10 USD per trade. On the measurement frame of 100,000 USD the run ends at 70,989.16 USD before our commission retrofit and at 46,765.32 USD after it. Nothing here needs interpretation. The run is simply negative, and it stays negative in every direction we push it.
The audit covers one tester run over XAUUSD on M1 from 2003-05-05 to 2026-08-19, with tick history rated 100% real ticks and a 10 ms execution delay parsed from the tester journal. The catalog protocol runs a fixed 0.01 base lot on a 100,000 USD frame at 1:5000, which are measurement settings that keep the tester from censoring the trade sequence rather than a recommendation. Trade pairing came from the journal with exact confidence and 0 chronology violations, and the reconstructed balance matches the report to 0.00 USD. The run is deterministic (0 randomizer prints), which matters because it means these numbers can be reproduced rather than resampled.
Four numbers, one ladder. Gross before costs, the run loses 29,030.55 USD. Swap is almost irrelevant at +19.71 USD because only 198 trades stay open overnight, which brings the tester native result to -29,010.84 USD, the figure the report head prints and the one behind the 70,989.16 USD end balance above. On top of that our commission retrofit of 3.50 USD per lot per side across 605,596 deals adds -24,223.84 USD, and that is the step from the head figure to the audited net of -53,234.68 USD.
A scalper that holds for a median of seconds pays the spread and the commission line, not the financing line. The sensitivity section puts a number on how thin the margin is. The break even shock is -0.18 USD per trade, so a cost world only slightly worse than ours changes the size of the loss, not its sign. At a flat shock of 0.50 USD per trade the profit factor falls to 0.351 and the run nets -204,633.68 USD.
The structure forensics flag is red, and not for lot escalation. The median size of an adverse add is 1.0x the open basket, so this is not a martingale. It is a grid. The EA added 42,534 entries against an open losing basket versus 32,744 with the move, up to 5 simultaneous positions, and it held long and short at the same time in 12,473 moments. Nearly half of all entries (46.5%) stack onto something already open. That mechanism is why the win rate looks friendly at 57.6% while the result does not. Winners close small and often, and the losing side waits for a rescue that the stop eventually pays for. The exit mix records 124,466 stop loss exits against 178,332 signal or time exits.
The product page records a release on 2026-06-02 and 3 updates, the last on 2026-08-14. That splits the tested history into three windows. Everything before the release, 8,419 days and 262,571 trades, was fully visible while the EA was built (net -49,821.53 USD, profit factor 0.78). The window between release and last update covers 73 days and 40,227 trades (net -3,413.15 USD, profit factor 0.86). The only slice that is guaranteed out of sample, the time after the last vendor update, is 0 days long in this run. Note the direction of the finding. The in sample window is the weaker one here, which is the opposite of the usual overfit pattern and is worth saying plainly.
The vendor recommends a deposit of 1,000 USD. Our survival map starts a fresh 1,000 USD account at every one of the 605,596 deal boundaries and replays what follows. 97.0% of those starts end in account death. Chained end to end, the same history burns 53 accounts before it reaches the present. On the prop side the picture is consistent. Under the IQ Capital Classic rules the drawdown sweep still records 1 death even at 1/8 of the tested size, with a withdrawal rate of 0.0% a year at every sizing step, so there is no size at which this history both survives and pays.
A bootstrap over 1,000 paths puts the probability of a net loss at 1.0, with a median profit factor of 0.785 and a median expectancy of -0.18 USD per trade. The median maximum drawdown across paths is 53.23% of the start balance, which is also what the end of day curve shows for the observed run (53.23%, annualised return -3.11%, end of day Sharpe -2.58). Twenty three of the 24 tested calendar years are negative, and the exception is 2018, which finished 1.30 USD up. Removing the 20 best trades changes the net result by 283.53 USD, which is the mirror image of the usual finding. There is no small set of hero trades holding this up, because there is nothing to hold up.
The challenge model closes the same door from the other side. Run against the FTMO rule set, phase one is passed on 0.0 of the 1,000 paths and both phases together on 0.0. What makes this different from a merely slow strategy is where the paths end. The probability of hitting the maximum loss floor before reaching the target is 1.0, and the probability of a path still running when the data ends is 0.0. Every single resampled path breaks the floor. There is no undecided case and no median time to pass, because no path ever passes.
The deal level census, the survival map and the prop rule replays need trade reconstruction, and that 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.