The window the vendor could not see: 13 of 26 audits have one at all

Gold Snap closed 3,769 trades in our run and exactly 20 of them closed after the date of its last update. Those 20 trades are the only part of the record that the author could not have looked at while building the product. They are 0.5 percent of the run, they cover 46 calendar days, and they made 29.73 USD. The other 3,749 trades ran over price history that was on the screen while the robot was being written and adjusted.

Every audit we publish cuts its trade history at two dates and reports the three pieces separately. This article reads that block across all 26 default audits and reports what it found, including the part that does not flatter the method.

Where the two cuts fall

The two dates are the release date of the product and the date of its most recent update. They produce three windows.

• Before release. Price history that existed before the product did. Fully visible during development.
• Release to last update. The product was on sale and was still being changed, so every update could see this stretch. The release day and the update day both belong to this middle window.
• After the last update. The only stretch that is guaranteed not to have been fitted. It starts the day after the recorded update and runs to the end of the trade history.

Four mechanical details decide what a row means, and each of them can move a number.

• A trade lands in a window by the time it closed. A position opened before an update and closed after it counts as out of sample. The count is generous in that direction and we do not correct for it.
• The window length is the overlap with the actual trade history, in calendar days. A length of 0 days means the window does not exist in this run at all.
• A missing update date is treated as the release date. No known update means everything after the release day counts as out of sample, which is the reading most favourable to the product.
• The dates come from the catalog build, out of our MQL5 product archive or an explicit override. A wrong date moves the cut without anything looking broken, which is why we print both dates in the audit next to the windows.

The three windows add up, and we checked it

The window figures sit on the costs level of the audit, after the commission retrofit, and not on the tester header. That is easy to verify by addition, so we did it on four runs.

• Gold Snap gives 1,725.16 plus 167.11 plus 29.73 which is 1,922.00 USD, the exact retrofitted result of a 101,922.00 USD final balance against a 100,000 USD deposit. Its tester header still reads 2,223.52 USD, because the 301.52 USD of retrofitted commission is not in that number.
• The Gold Reaper gives 2,706.75 plus 1,161.83 plus 173.25 which is 4,041.83 USD against a final balance of 104,041.83 USD.
• Smart Gold Impulse gives 779.74 plus 0.00 plus 146.46 which is 926.20 USD against 100,926.20 USD.
• Twister Pro Scalper gives 239.82 plus 87.64 plus 2.26 which is 329.72 USD against 100,329.72 USD.

Thirteen of twenty six runs have such a window

The block behaves in four different ways across the catalog, and the three ways that are not a real window matter more than the one that is.

• 13 runs carry a window with at least one day and one trade. They are the subject of the rest of this article.
• 7 runs carry the block with a window of 0 days and 0 trades. Lizard and Ultimate Breakout System were both updated on 2026-08-28, Logan on 2026-08-26, Quantum Titan and Scalping Robot Pro on 2026-08-25, Smart Gold Hunter and Range Breakout EA with Range Filters on 2026-08-23. Every one of those dates falls after the last trade of the run, so there is nothing behind the cut. An update published after a backtest ends erases the out of sample window of that backtest completely.
• 2 runs carry the block with the plain string no release date. OilVector X and Quantum Queen have trades and a window, and we have no date to cut them at.
• 4 runs carry no market block at all. Market Anomalies EA, Pulse Engine, Quantum Emperor and Waka Waka. The engine writes the block only when the catalog build hands it at least one market date, so their audits are silent here rather than wrong.

How big the window is when it exists

Across the 13 runs that have one, the median window is 46 days and the median count is 47 trades. The spread around that median is what a reader should carry away.

• The largest share of any run belongs to Smart Gold Impulse at 268 of 4,163 trades, which is 6.4 percent. It is the only product in the catalog whose recorded release and update dates are the same day, so its middle window is one day long and holds no trades.
• The longest window belongs to CryonX at 201 days and 173 trades, 4.2 percent of its 4,127.
• The smallest share belongs to Twister Pro Scalper at 2 of 986 trades, which is 0.2 percent over 5 days. ThunderGold Scalper sits next to it with 3 of 984 trades over the same 5 days.
• Eleven of the thirteen stay under 2 percent of their own run. Gold Snap at 0.5 percent, Adaptive Gold Scalper at 0.9 percent over 24 days, Gold Trade Pro at 0.9 percent over 84 days and The Gold Reaper at 0.9 percent over 78 days are all in that group.

Six get better, six get worse, one cannot be compared

The obvious thing to do with this block is to compare the profit factor before release against the profit factor after the last update. We did, and the result is not the tidy story an audit site might prefer.

Six read worse out of sample. CryonX falls from 2.62 to 0.79 and is the only one of the thirteen that is net negative in the window, at 59.52 USD in the red. Gold House falls from 2.02 to 1.37, ThunderGold Scalper from 1.54 to 1.09, Gold Atlas from 1.35 to 1.07, Prop Firm Gold EA from 1.25 to 1.05 and Gold Snap from 1.66 to 1.62.

Six read better. Quantum Athena X rises from 3.93 to 11.46 on 47 trades in 11 days, The Gold Reaper from 1.92 to 9.65 on 16 trades, Quantum Queen X from 4.30 to 7.50 on 43 trades in 11 days, Gold Trade Pro from 1.71 to 2.44, Adaptive Gold Scalper from 1.19 to 2.16 and Smart Gold Impulse from 1.14 to 1.52.

One cannot be compared at all. Twister Pro Scalper has two out of sample trades and neither of them lost, so there is no denominator and the field is empty rather than infinite.

The split is six against six, and that is the finding. A profit factor of 9.65 built on 16 trades is not evidence that a robot works after its last update, and a profit factor of 0.79 built on 173 trades is not proof that one fails. What this block delivers is a sample size next to a number, so a reader can see how thin the only unfitted evidence really is before deciding what it is worth.

The same run, two different answers

One pair in the catalog makes the dependence on metadata impossible to miss. Quantum Queen and Quantum Queen X are the same measured run published under two slugs, and both count 3,357 trades. Queen X carries a release date of 2026-07-14 and an update on 2026-07-31, and reports 43 out of sample trades over 11 days. Queen carries no release date and reports nothing. Identical trades, two different answers, because this block reads dates and not deals.

Run these checks yourself

• Ask for the release date and the last update date before you read any backtest. Without them a curve over 23 years and a curve over 23 days look the same on a chart.
• Count the trades after the last update, not the days. Two hundred days of a quiet strategy can hold fewer trades than five days of a scalper.
• Check whether a fresh update just erased the window. Seven of our own runs sit at zero for exactly that reason, and it will happen to every rerun we publish.
• Add the windows and compare with the net line. If the three pieces do not sum to the result, some of them are on a different cost level.

Every figure above is public in the audit files linked in place, and the category pages list the verdict lamps for free. Your own tester report? The browser check is free.