A fair question we get asked: how do you know the AI really tested anything? Here is the honest answer. The AI does not test. It turns your words into a rule, and then plain arithmetic counts what that rule would have done on real candles.
Last updated 16 September 2026. The worked example below was run on that date.
The AI translates. You describe a strategy in your own words, or paste a video link. The AI turns that into a rule: entry conditions, where the stop goes, where the target is. It then shows you that rule in plain English, and lists the parts it could not test. If it read you wrong, you can see it before you believe any number.
The engine counts. The test itself has no AI in it. It is arithmetic over real candles: find every signal, open the trade, walk forward candle by candle, record where it ended. Run the same test twice and you get the same answer, because nothing in it has an opinion.
The test uses actual price history from Deriv. On a 15 minute chart it takes up to 20,000 candles, which is about seven months.
The most recent 15% of the history is locked away and not used. That is saved for the final check, so there is always some price action the strategy has never been near.
The first part is called the older days, the second the newer days. Nothing is shuffled. Markets move through time, and mixing the days up would let the test see the future.
Every time your conditions appear, the trade is taken. No picking the good ones. The stop goes where you said. The target goes where you said, or at 2:1 if your strategy never names one.
For every candle after entry, the test asks whether the high or the low reached your target or your stop. If one candle touched both, it counts the stop. That is deliberately the pessimistic assumption. If neither happens in time, the trade closes at the time limit.
Risk is the gap between your entry and your stop. Scoring everything against it means a test on Volatility 75 and a test on Boom 500 can be compared on the same scale, whatever the price of the market.
This is the point of the split. Any rule can look brilliant over one stretch of history, especially the stretch someone picked for a video. Making money over two separate stretches is a much harder test to pass by luck.
Run once, on the newest slice from step 2. This is the closest thing to forward testing without waiting months, and it is where most strategies fall over.
This is not an illustration. On 16 September 2026 we pulled real Volatility 75 candles and ran our own desk strategy, the sweep and structure shift setup, through the tester. Here is everything the engine recorded.
The strategy made money on the older days and again on the newer days, so it passed the split test. Then it lost money on the month it had never been near, badly. The verdict is failed, and the final check is the only reason we know.
This is our own strategy, published here failing. We would rather show you that than a screenshot of a good week.
Every trade is listed in the app, with the time it opened, the direction, the entry, the stop, where it ended and why. These three are real rows from the run above. You can put any of these times on the chart and check them yourself.
| Opened (UTC) | Side | Entry | Stop | Ended at | Why it ended | Result |
|---|---|---|---|---|---|---|
| 2026-02-22 21:15 | Short | 34,651.6 | 35,079.3 | 33,796.3 | Target reached | +$200 |
| 2026-02-24 03:00 | Long | 36,373.8 | 35,856.0 | 35,842.5 | Stop hit | -$103 |
| 2026-06-06 14:45 | Short | 30,274.8 | 30,622.0 | 30,790.5 | Stop hit | -$149 |
Notice the last two lost more than the $100 that was risked. A candle can jump straight through a stop level, and the test counts what the next available price actually was rather than pretending you got out exactly at your stop. Real losses behave the same way.
In the test above, 42% of trades won and the strategy still made money, because the winners were bigger than the losers. A system that wins 80% of the time and gives it all back on the losses is worse. Judge the money per $100 risked, then look at the win rate.
A strategy with no real edge passes the two halves by luck roughly one time in four. That is why passing gets called a candidate, never a winner, and why the final check exists.
If you change the stop, re-test, change the target, re-test, and keep going until something passes, you have not found an edge. You have found the version that fits this history. The app counts how many times you have tested on the same market and says so on screen from the fifth test.
Paste a strategy in your own words, or a link to the video you saw it in. You will get the rule the tester read, the parts it could not test, the two halves, the final check, and every trade it took.
See the real chart, no sign-up