The backtest that predates the product: 20 audits split at their release date

Gold Snap was published on 2026-04-11 and our run of it covers 8,498 days of gold history. Of its 3,769 trades, 3,720 closed before the product existed. Another 29 closed while the vendor was still shipping updates. That leaves 20 trades in the only stretch of the record whose data arrived after the last change we know about. Every audit we publish carries this split, and it is the cheapest question a buyer can ask of a track record.

Where the two cuts fall

The engine cuts the trade list at two dates taken from the product page, the release date and the date of the last update. Three windows come out of it, and each one answers a different question.

• Before release. The price history existed and the product did not. Every candle in this window was available to whoever built the robot, in full, for as long as they wanted.
• Release to last update. The product was live and still being changed. Each update could see the data that came before it.
• After the last update. The only window whose data arrived after every change we know about. It is the one figure in the block that carries weight, and it is usually the shortest.
• A trade lands in the window it closed in. The cut runs on exit time, so a position opened before a release and closed after it is counted in the later window.
• The release day and the update day belong to the middle window. Where a product has no recorded update, the release day stands in for it, and everything after that day counts as out of sample.
• Window length is printed next to every figure. A strong result over five days is not evidence, and the reader has to be able to see the length without asking for it.

How much of a backtest is older than the product

In all 20 audits that carry the split, the majority of trades closed before release, and the smallest majority is still four out of five. Range Breakout EA is the mildest case with 1,235 of 1,525 trades. Quantum Titan is the extreme with 1,940 of 1,947, which leaves seven trades from the whole run in the two windows after publication.

The reason is arithmetic rather than suspicion. Fourteen of the 20 products were published in 2026, three in 2025, two in 2024 and one in 2023, while our gold history starts in 2003. A backtest that runs to the beginning of the data will always be mostly older than a product released last spring. What the split adds is the number, so nobody has to guess how much of a headline figure comes from a period the builder could inspect at leisure.

The window the vendor could still see

Eighteen of the 20 audits carry a profit factor in both the before release window and the release to last update window. In 15 of those 18 the middle window is the better one. The three exceptions are Gold House at 2.02 against 1.51, Quantum Queen X at 4.30 against 3.04 and Quantum Titan at 1.44 against 0.84.

Some of the gaps are large. Gold Snap reads 1.66 before release and 17.91 after it, CryonX reads 2.62 against 16.90, and Logan reads 0.26 against 3.10. Four of the 20 lose money in the window before release, and three of those four show a winning profit factor in the middle one. Logan is the sharpest example we hold, with 24,807.11 USD lost across 1,249 trades before its listing and 472.46 USD earned across 123 trades in the 49 days after it.

The size difference is the first thing to say about that comparison. The median middle window in this set is 87 days long, against before release windows that run into thousands. A profit factor computed on 123 trades and one computed on 1,249 are not the same kind of number, and the block does not pretend otherwise. What the count of 15 out of 18 shows is a pattern worth naming, not a mechanism we have identified.

Seven audits have no out of sample window at all

In seven of the 20 the third window is zero days long. The cause is mechanical and it is the same in all seven. The product was updated after the last trade in our run, so there is no history left on the far side of the marker.

Ten to eighteen days is the whole story. These are not abandoned products and it is not a defect in the run. It is what happens when a robot is maintained faster than a measurement can be finished, and it means the honest answer for those seven is that we have no unfitted evidence at all rather than weak evidence.

The thirteen windows that do exist

Across the 13 audits that have an out of sample window, it holds 1,155 trades in total. For comparison, Gold Atlas alone closed 10,744 trades before its release.

Twelve of the 13 are profitable out of sample, and the one that is not is the longest window in the set. CryonX ran 201 days after its last update and lost 59.52 USD at a profit factor of 0.79. The median window here is 46 days and the shortest two are five days holding three trades and two trades. Read the table from the left and the pattern is plain, because the largest profit factors sit on the shortest windows.

The last row is worth a sentence of its own. TwisterPro Scalper has no profit factor out of sample, and the field is empty rather than low. A profit factor is gross profit over gross loss, and both of its two trades won, so the divisor is zero. Empty means no losses here, not no data, which is exactly the kind of field a summary table silently gets wrong.

What the split cannot say

The two dates are not our measurement. They come from the product page through our daily watcher, and all 20 splits in this article read the archive without any manual override. A page that gets republished carries the date of the new listing, which is why Quantum Queen X shows a release in July 2026 for a strategy that was listed and delisted before. A wrong marker moves both cuts and nothing in the run looks broken afterwards.

Out of sample means after the last update we recorded. If a vendor changes a product without the page noting it, the third window still gets printed and it is no longer unfitted. The block is only as honest as the update history behind it.

Six of our 26 default audits carry no split. Two are marked as skipped for want of a release date, OilVector X, whose scraper archive is missing, and Quantum Queen. The other four are older runs that predate the block, Market Anomalies EA, Pulse Engine, Quantum Emperor and Waka Waka, which are unlisted and which we name without linking. That also means the 20 audits counted here are 20 distinct runs, because Quantum Queen and Quantum Queen X are the same measured run under two slugs and only one of them carries a split.

And a window is not a forecast. Twenty trades over 46 days decide nothing in either direction, which is why our own audit page prints the length in the same sentence as the result. The split tells you how much of a record was inspectable while it was being built. It does not tell you what the robot does next.

Run these checks yourself

• Read the window length before the profit factor. A figure of 9.65 over 16 trades and a figure of 1.07 over 215 trades are not comparable, and the larger one is the weaker claim.
• Ask when the product was first listed. If the answer is months ago and the chart is decades long, the chart is a simulation of a product that did not exist for almost all of it.
• Ask when it was last updated. An update that is newer than the data means there is no unfitted evidence at all, and that is a different sentence from weak evidence.

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.