Since most EAs have history reader, who should i contact at MQL5 to complain about the state of this marketplace ? - page 4
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Yes I discussed with Quentin, it was painful. He was impolite, offending and arrogant, though I finally got a report (4 pages PDF about one EA "statistical" analysis).
The "so easy to prove that many EAs" is actually reports done by an AI agent. I read it and I investigated a bit. There is NOTHING in that report that proves anything. It just show, clearly, that from the publication date, the EA was behaving differently compared to the historical period. It doesn't prove anything about the cause. The most probably cause is just over-optimization, curve fitting on history data, for sure you will find a lot of such EAs on the Market and everywhere. Cheating is still possible, as all is always possible unless proven otherwise, but proving it is not so easy and should not be taken lightly.
AI agents used by people who don't understand what they are talking about lead to such strong affirmation as "most EAs have history reader" or similar things. The only thing which was proven was : AI agents used by incompetent people is dangerous, and lead people to overestimate themselves and theirs "findings".
You need the custom symbol to shift the dates in order for the EA to trade in the cencored periods. If the result is not the same, this is evidence of manipulating through history. And yes, there are such EA in the top list at the market.
I also investigated that case (an other EA), and while there are some interesting information, there is also no proof of causality about cheating. What is proven is a correlation not causality. There are legitimate usage of hardcoded data (dates) in EA for the Market, finding dates in an ex5 prove nothing. Using a custom symbol with shifted dates, and having different results compared to the original, is also not a causality proof of cheating.
Again, I am not saying there is no cheating this or that EA, I am just saying the evidences are weak or insufficient.
Yes I discussed with Quentin, it was painful. He was impolite, offending and arrogant, though I finally got a report (4 pages PDF about one EA "statistical" analysis).
The "so easy to prove that many EAs" is actually reports done by an AI agent. I read it and I investigated a bit. There is NOTHING in that report that proves anything. It just show, clearly, that from the publication date, the EA was behaving differently compared to the historical period. It doesn't prove anything about the cause. The most probably cause is just over-optimization, curve fitting on history data, for sure you will find a lot of such EAs on the Market and everywhere. Cheating is still possible, as all is always possible unless proven otherwise, but proving it is not so easy and should not be taken lightly.
AI agents used by people who don't understand what they are talking about lead to such strong affirmation as "most EAs have history reader" or similar things. The only thing which was proven was : AI agents used by incompetent people is dangerous, and lead people to overestimate themselves and theirs "findings".
I also investigated that case (an other EA), and while there are some interesting information, there is also no proof of causality about cheating. What is proven is a correlation not causality. There are legitimate usage of hardcoded data (dates) in EA for the Market, finding dates in an ex5 prove nothing. Using a custom symbol with shifted dates, and having different results compared to the original, is also not a causality proof of cheating.
Again, I am not saying there is no cheating this or that EA, I am just saying the evidences are weak or insufficient.
So you will leave it there right ? Everybody can make their own conclusions.
Note : I did not criticise you at any point publicly but you chose to criticise me in this forum.
If I was impolite, offending and arrogant then you are uninquisitive, ignorant and blind.
Also note : You wrote this comment without even coming back to me though our private messaging on Telegram - which is extremelly un-professional.
So to summarise :
*My objective was to tell you that a lot of EAs on this markets use backtests cheats.
*Your objective was to shut me down. And to do so - you choose to hear, see and do nothing.
I think we are done here. I now have made my opinion on you and the efforts you are (not) willing to make.
But I'm not yet willing to let go.
I have sent Alain the full breakdown of the evidence in Excel.
This one has been written by me.
Please review and let me know what you think.
This is as much as can be analysed without studying the code.
But I'm not yet willing to let go.
I have sent Alain the full breakdown of the evidence in Excel.
This one has been written by me.
Please review and let me know what you think.
This is as much as can be analysed without studying the code.
I am suggesting you to post this Excel file here, as long as the EA and/or the seller is not identified, there is no problem to post such report here. So everyone can check it and contribute.
I already provided my conclusion : it proves nothing in terms of "backtests cheats". At best it proves the EA is trading differently from August, which is the time it was released.
But I'm not yet willing to let go.
I have sent Alain the full breakdown of the evidence in Excel.
This one has been written by me.
Please review and let me know what you think.
This is as much as can be analysed without studying the code.
Sure, good idea. It's there for everyone to see then :)
This file shows the performance history of a yet unnamed EA on the MQL5 marketplace, from a standard MetaTrader 5 backtest on gold (XAUUSD) from January 2018 to October 2026: 1,479 trades, all at 0.01 lot.
This EA does not use Grid or any sort of recovery trade.
* Up to 30 July 2026 the history looks too perfect: 100 of 103 months profitable, a maximum drawdown of 1.24%, and a Sharpe ratio of 33.9. As a simple comparison, no human or machine can ever achive an information ratio above 5.
* Over those eight and a half years, its typical winning trade was about 1.7 times its typical losing trade, every single year. Winners were never once smaller than losers; the lowest ratio was 1.04.
* From 31 July 2026 the pattern flips: winners become less than half the size of losers (0.43), and the EA trades 50% more often.
* Any statistician will demonstrate that this change of character is simply impossible in real life as it divides performance in two perfect "character" camps. Sadly, a genuine strategy does not change character on a fixed calendar date, but a replayed list of known winning trades does when the list runs out.
* Ironically, the EA still made a small profit after the change (557$ without cheat vs 3532 with cheats on - ouch !), but its honest track record since the 30th of July is only 53 trades long.
* This EA - and so many others in this market - are destined to fail. No problem ! We already know the author will simply strategically plull the EA from the market and not suffer any consequences for their action.
Hi. I could create an EA for you that shows tens of billions percents in profit by July 30th. That doesn't mean the EA is actually reading historical data. And you can't read historical data without having some form of pre-existing data or a data file. The real issue is that very few brokers maintain high-quality historical data over long periods. That’s why, for backtesting, you should generally only consider the last few months and only if the backtest quality is 100%; anything less, like 99%, is already not good enough! The MT5 tester is exceptionally good, but it has to be used intelligently. Any issues with historical data lie with the brokers, not with the MT5 tester itself.
P.S. This "issue" has been discussed in several places on the forum; it might actually be very useful to create a dedicated thread for it, as there is a clear, widespread lack of information among many users.
P.P.S. The only thing MetaQuotes could do is prevent the MT5 tester from running in "every real tick" mode for periods where data quality is not 100%. It should only allow "every real tick" testing for periods where the quality is 100%, not a single percent less. That is, after all, what "real tick testing" implies; however, due to the historical data provided by the broker, tests with a quality of less than 100% are no longer true "real tick" backtests. For those periods, the tester should default to other testing modes.
Hi. I could create an EA for you that shows tens of billions percents in profit by July 30th. That doesn't mean the EA is actually reading historical data. And you can't read historical data without having some form of pre-existing data or a data file. The real issue is that very few brokers maintain high-quality historical data over long periods. That’s why, for backtesting, you should generally only consider the last few months and only if the backtest quality is 100%; anything less, like 99%, is already not good enough! The MT5 tester is exceptionally good, but it has to be used intelligently. Any issues with historical data lie with the brokers, not with the MT5 tester itself.
P.S. This "issue" has been discussed in several places on the forum; it might actually be very useful to create a dedicated thread for it, as there is a clear, widespread lack of information among many users.
P.P.S. The only thing MetaQuotes could do is prevent the MT5 tester from running in "every real tick" mode for periods where data quality is not 100%. It should only allow "every real tick" testing for periods where the quality is 100%, not a single percent less. That is, after all, what "real tick testing" implies; however, due to the historical data provided by the broker, tests with a quality of less than 100% are no longer true "real tick" backtests. For those periods, the tester should default to other testing modes.
We are talking about active cheats within the code here. Statistically demonstrated.
Sure, maybe as a developper you don't cheat and that is great :)
You got nice EAs man :) https://www.mql5.com/en/users/tumcog/seller
We are talking about active cheats within the code here. Statistically demonstrated.
You are not identifying the problem correctly. The issue lies with the historical data. If you could show me a backtest with 100% historical data quality, where the EA executes exceptional entries, then perhaps you would be right. But no such thing exists. And as I mentioned in my previous message, I could create an EA that shows even more exceptionally good results over past years, without it actually reading any data at all.
Oh I am identifying the problem very correctly given the number of people contacting me :).
But for your thirst of knowledge, here are the stats with 100% qual data. Your query sorted.
Note that the statistical chance that this sudden change is random is 0.17%.
"In the land of the blind, the one-eyed man is king"