AI ORASignal - First test results (we are always in profit)

Again #CHFJPY
Today is a classic ORA example:
perfect entry + solid reserve and a clear edge.

This pair has clearly caught the AI’s attention, and that alone is already valuable knowledge.
Why does a certain pair start regularly attracting the AI (especially one with such strong stats)?
Another interesting approach for those who trust themselves first, but are still open to considering powerful signals from the system


PS While we were preparing this post, the AI took another lot on this pair. It’s still open.
Interesting… will it be right again?

First interesting case, so we’re going a bit live — showing how we act in real time.
(mode f30)

We’re currently stuck on #CHFJPY.
To be fair, it was expected: all-time highs again + geopolitics. We could talk about it for a long time, but there’s little point.

However, we trust the AI.
If she found these levels, they are not random for her. Knowing how she works, we take that into account.
That’s why we’re adding our own manual lots in the same direction as the AI’s lots.

We’re not closing them.
We believe in her ))

Let’s see how it ends. We’ll be waiting

Continuing the #CHFJPY story.

As we said — we’re waiting and trusting the AI

Interesting moment: isn’t this starting to look like a “Head and Shoulders” pattern? )))


With our AI we have a different kind of intuition. Not retrospective, but statistical.
We wait, we don’t close, but only while waiting does something become clear — something that many people might recognize later, but which wasn’t visible or even hinted at before.

Let’s see how it plays out, and later we’ll add some commentary.

Its climbing instead falling for 3 pips tp

First important note — we are currently at some all-time highs on several pairs.
For example, #GBPCAD was last at this level back in January. On some pairs we’re literally at absolute historical highs.
Why does this matter? Psychology. What used to feel calm suddenly becomes much more tense. A lot of new variables appear literally out of nowhere (emergence).

Regarding “it’s climbing instead of falling for the 3 pips TP”:
Everyone has their own intuition.

But one thing is clear — price movement is not a simple linear function that just goes down. It’s more like a tug-of-war, which we’re watching in real time.
All we can see is a 2D slice of an n-dimensional space.

We already had the thought that the levels found by the AI should, in theory, produce a bounce. After all, it’s a trend change, which means stops, liquidity grabs, and other unexpected moves.
However, trading against the AI’s direction in hopes of a bounce is usually a weak strategy.
Just look at most of what we’ve posted before.

In any case, all of this is just interpretation. We’re still looking at a chart that only shows price in a thin slice of time.
I’ve previously posted examples of pure random charts. Even there you get the feeling “it should go down, but it goes up”

I don’t know.
We look at the market differently. Different intuition

Continuing the #CHFJPY story
A really interesting case on this pair during these tough days.

Straight away — the AI has already evolved and now understands such situations much better. The “bad gene” was found — a very loud one that was screaming something was off from the very beginning — and it has been fixed.

However, the trades were already taken.
The AI lot (+3 of our manual ones) were placed and they were statistically justified.
From the chart, it’s clearly visible how the bears spent the whole day smashing against the same level — a level that even broke a classic, already-formed “Head and Shoulders” pattern.
In short — we decided to hold them longer
Come what may.

After all, there’s only one lot from the AI here, so formally it doesn’t ruin the statistics.
The rest is our experiments and a bit of excitement

If you’re trying to copy her trading, here are a few tips:

First — look at the entry price and preferably wait for an even better position (it often takes just 2–3 minutes).
If it’s a short — wait for it to bounce a bit higher and then enter. If it’s a long — do the opposite.


Secondly, there is the entry point and its expected TP. Think of it as your aiming mark.
Beyond the TP — it’s complete unknown territory. And if the price has crossed the TP even once after entry, re-entering after that is strictly forbidden )))
And if you don’t want to dive deep into the currency pair’s situation, then as soon as the price reaches or exceeds her TP (or if you entered from a better price, as mentioned above, and are already in profit), it’s usually better to close the position early and not wait for further trend development.
That’s already an additional process that requires attention.

And of course — try it on a demo account first

And one more thing from me personally:
I’ve been copying her too. If the price is very close to her entry — I usually don’t jump in. But if it bounces and gives extra room — then I enter.

Hope you like it

#CHFJPY — positions closed.
In the end, we fell just a little short
Let’s forgive it this small slip-up (which is still extremely rare for it)
A small loss of a few percent on the AI f30 mode.
We try not to hold lots longer than one day — especially when price has reached the all-time high not by accident.

But the main thing: at the end of the day the AI took two lots in the opposite direction, which still haven’t shown any reserve yet (we expect a stronger bullish trend again tomorrow).
Based on her track record — the trend has already changed, and holding now feels pretty sketchy

New day, new events. This one is done.

Someone might say: “Come on, trading like this when it’s unclear and you occasionally catch some nasty MAE…”
We agree!
Don’t trade like this
Everyone builds their own statistics anyway

Our approach is simple.
We have the AI (already with a solid statistical history) and she calls the shots here.
Why? How? God knows (well, we know what’s under the hood, but that’s beside the point).

And by the way, I personally couldn’t show such a win rate and profit factor month after month )
Who can? We’re genuinely happy for them

So we’ll keep trusting her and moving forward

Sorry I missed your question earlier.
By the way, it’s a really good one

The system can be scaled to longer timeframes, which of course can give much more pips.
I can’t explain everything, but in order for the models to actively evolve, we need high dynamics and speed.
That’s why we scaled them down to the smallest intervals, basically turning them into a scalper automatically.

But I’ll emphasize once again: there are no perfect boundaries. How much reserve you’ll get needs to be monitored.


And one more important point that many people miss.
I’ll share my personal opinion: the desire to “catch bigger moves” contains quite a few traps.

Psychologically, one big trade always feels much easier than a bunch of small ones (only over the long term does the average pips per trade make “catching bigger moves” seem worthwhile).
But I think this is exactly where most of the problems come from.

Trying to see from time t0 into time t1000 is still an incredibly difficult task — and in many cases, physically impossible.
Even the most precise weather forecasts, no matter how hard meteorologists try, only remain reliable for about 14 days ahead (the limit of practical predictability of detailed weather) )))

From the point of view of an engineering and scientific approach, any statement must possess the main quality — repeatability under similar conditions.
In other words: if a system states a hypothesis and forms an expectation, then upon repetition of similar market circumstances this expectation must be realized with an acceptable statistical error. If this does not happen on a large sample — the method is considered non-working.

Yesterday ORA AI placed two longs with an expectation of growth. The growth occurred — moreover, classic, smooth, increasing, with a clear change of the previous bearish resistance.
We closed the shorts in a timely manner.

Interestingly, before the AI signals the bearish trend really looked strong. That is why the fact of the direction change became a particularly valuable confirmation: the robot caught exactly the regime shift, and not followed an already obvious movement.

Looking at the chart as a 2D slice of a complex stochastic process P(t), we see not a “linear function”, but dynamics similar to a change in heart rhythm or a tug-of-war between the volumes of participants.

And between the lots from the AI it seems to us — the “heart rhythm” is clearly different )

We personally do NOT expect a “beautiful” linearity of the trend.
And it is precisely here that many begin to develop magical thinking.

Science, however, requires something else: not to believe in ideal predictability, but to measure how stably the system works on large statistics.
So far our statistics say that the system is exactly doing this

Here are a few more interesting cases.

For example, yesterday there were lots on these pairs.

This is how I look at what is happening, analyzing it retrospectively.

Yes, everything is very simple – would I have blown the deposit or not. Would it have been worth waiting or not if it had gotten stuck.
Would I have had a second and third chance, or at least an option to minimize the loss, if I had waited?

These are the questions I ask myself when looking at the lots from the AI.
Time after time I see that blowing the account with it, de facto, is really very difficult

For example:
GBPJPY in the example – it closed with a minimal margin, but would it have been worth waiting if it hadn’t closed at all? Trusting the AI again? And looking again – yes, it definitely would have been worth it in this case too.
GBPCAD in the example – there were several opportunities to close in profit and now it is clear – the bears are still strong.

And this is on the currently most difficult pairs. For example the others, GBP/USD – there everything was excellent right away.
And pairs with CAD or CHF, JPY remain under strong pressure due to oil and as safe-haven assets.
And here the issue is not with the calculations, but with a sharp increase in liquidity, stop triggers and other circumstances that make the processes exceptionally random (unpredictable beyond the boundaries of chance) and, by the way, that is why such a test for the AI is excellent, on all pairs, without exception.

If the little machine manages to be universal

PS:
And I know there will be those who will say – "this is some kind of magic and it’s far-fetched".
I personally tell them – these words, these thoughts – are confirmed by statistics.
Your words, your thoughts – are confirmed by your statistics.
Therefore, it’s not thoughts that fight, but statistics, while we share thoughts and something truly new, again, if we follow the statistics )

Once again – we looked at it differently, from a different angle, and showed the real result. We are perfectly familiar with the previous views and angles. They are here in every thread. Even on this forum. Only for some reason no one attaches accounts with statistics.
That’s why this different view might actually carry some weight, and it’s not worth dismissing something that is, de facto, RADICALLY different from the rest )

A bit more about CHFJPY, where that rare “loss” from ORA AI occurred (rare strictly in statistical terms).

What does a 95% win rate actually mean? It means that out of every 100 trades, 95 are correct and roughly 5 may raise questions.
Given that the AI places dozens of lots per week, these rare 5 trades are bound to appear from time to time.

And what interests us is exactly what these trades look like.

Studying the 95 successful ones is not particularly interesting. We are much more interested in the rare and conditionally “bad” lots.

In this n-dimensional system, there is no such thing as absolutely good or bad — everything has some dispersion and variance.
So even a difficult trade — what is it really? Would it have gone against us and never returned? And if it did return, how often and how quickly? In the next similar situation, should we wait and trust the statistics, or is it better to close the position?

Here’s what we see on this pair (after about 24 hours):
In fact, everything is quite simple.
Essentially, if we hadn’t panicked at all, these lots would have been closed in profit within a day.
In other words, these “bad” lots, which were experiencing pressure from the counter-trend, were found by our AI in places where even a strong bull could not break the bear that the AI had identified (or vice versa).

Behind all this philosophy lies the main thing we keep testing over and over again:
Would I have gone into a loss if I had trusted the AI or not?

So far, there hasn’t been a single case where we suffered any real or noticeable losses by trusting it.
They simply don’t happen — at least not up to this day and for many months now.
And we continue to monitor the situation after the event very closely, which pleases us no less

You see, even this single case over the last couple of months still doesn’t give us any peace

The second extremely unsuccessful trade from our AI this week )
After months of excellent performance.

Well, yes — this week is really tough )))

The losing trade is now on mode f9.
The other mode (f30) performed excellently today.

There were many successful trades on both f30 and f9, and only one, opened at 14:16 UTC, turned out to be extremely unsuccessful.
It looks terrible, but it’s important to note that only one lot came directly from the AI. All the others are our manual additions, made according to our overall tested logic and hypothesis (that such trades can ultimately bring even greater profit).

Obviously, we will wait until tomorrow

We respect and trust the AI’s opinion (we’ve spoken about this a lot before).

Everyone trades differently, but the key question remains the same: to hold or not to hold.

Even with our 95% win rate, out of every 20 trades there will always be 1–2 difficult ones (nothing in this world is 100%).
We personally need a clear and precise understanding of how to behave in such situations.
This is exactly what we are testing.

Just like in the previous case — we need to wait and trust.

Evaluations, reasons, and interpretations are secondary.
(Past trade statistics come first)


PS
Speaking about this, we actually touched upon the key reason why so many people blow up their accounts in trading.
Even if you do everything right 90% of the time, there will always be difficult cases.
The strategy for handling exactly these situations is what stabilizes and grows the deposit in the long run.
That’s why we personally love such situations.
And the AI itself has already “been slapped on the wrist” for this mistake (and has become smarter because of it).
It won’t repeat it — especially since the other AI lots on the same pair today went through successfully.

We continue discussing that one unsuccessful AI lot + the series of additional trades placed manually according to its logic

I’m sure many people are interested — or at least dream of a system that can confidently and consistently grow profit, and only AI is truly capable of this (why and how is a whole different big question).

That’s why we believe this topic is interesting to many.
After all, a trader is always on the side of strategies and, by extension, on the side of our AI — rather than brokers or Nature, which can sometimes be very unpredictable

That said, let’s remind ourselves once again: the physics of the Universe does not allow for 100% certainty. There is nothing that is 100%. This is not just big words — it’s a scientific fact.

Our task is to solve the problem of these rare “difficult” lots.

Let me remind you: even with a 90–95% win rate, the remaining percentage consists of these “difficult” cases (roughly 1 or 2 out of every 20 trades).
These are exactly the ones we focus on, because they are the most interesting. Moreover, every “difficult” trade has its own dispersion — they are all different. They are essentially outliers from the general flow of good statistics.

That’s why a difficult case is incredibly interesting, and this is what we’re talking about here.

We had one difficult AI lot. For the AI itself, it doesn’t spoil its statistics, and by the way, it has already evolved from this experience.
Right now, our additional manual lots are the ones affecting the statistics (greed? Risk management error?), all within a certain hypothesis that we are currently testing.

But the most important thing is to understand the nature of a difficult case: when you have strict and proven statistics behind you, and suddenly an outlier occurs.
Let me emphasize — this is not based on an assumption of what “might” happen, but on the AI’s strict statistics. The AI understands the market, it doesn’t guess (and this is also strictly measured by its metrics).

So if this is the case — what exactly are these lots?
It still seems to us that this is merely a delay in time of the expected result.
On the chart, we marked this “tug-of-war” between bears and bulls (the lot was on the bull side, long).
It’s important to remember that chart graphs are not a linear function. At every point in time, it’s simply the current balance of power between bears and bulls.
The bulls are clearly not giving up, and we even managed to close one lot in profit.

In the end — we continue to wait
By the way, not even 24 hours have passed yet.
That’s not a long time at all


PS
By the way, if the number of difficult lots starts to grow more and more, that would be a clear sign that the system is falling apart.
At the moment, there are no such signs.
This is clearly visible from the statistics — even yesterday there was only one unsuccessful trade, while there were more than ten successful ones.
But yes — anything is possible. Everything that can happen, can happen )))
However, so far there are not even small signs of this.
Definitely not today )))

CHFJPY — a great series of shorts and longs (a hook from the left and from the right for the market).

This was on the AI mode f30 just recently.
By the way, the same happened on mode f9 as well (on the backup account, and its statistics are also displayed on the main page of ForexFactory).

All lots are purely from the AI.

In my opinion, it looks beautiful



By the way, another aspect of the 90-95% probability versus the 5-10% of “difficult” lots.

It’s like a casino in reverse.

Every time I enter a trade, I have a 90-95% probability that the lot will be successful

But that’s already games and tricks.
What we find much more interesting is real stability WITHOUT any roulette — even with such a high probability.
It always delivers a far more confident and reliable result.

Beautiful night lots (UTC time) on perhaps the wildest pair out there — GBPJPY )

Once again, after some time has passed, you can clearly see the precise and accurate forecast for the trend reversal.
In other words, the trend break clearly happens right after the AI’s entry point. It can be either very obvious or less obvious.

By the way, yesterday’s CHFJPY, a few hours later — that’s exactly the moment when the AI throws a double hook at the market )
It places a sequence of lots in opposite directions of movement. When we see the AI make such a decision, we expect the boundaries to expand (the standard deviation on the current pair will widen).

The latest GU signal the profit very far far away from the target may I know why your AI put a very slim target TP

Thank you for your attention and for this question

I’ve already answered it before, but let’s go into a bit more detail this time.
I think others might find it interesting as well.

I assume you’re referring to EURJPY (judging by the timing of the question and the signal that was posted).
In the screenshots, you can see several other trades from today, including EURJPY (the purple line marks the Iran-related news. You can also clearly see from the volumes that the market “woke up” quite noticeably).

Also on the screenshot is the standard deviation in points across currency pairs — let’s take a closer look at that.

Pay attention to the first three columns.
They show how many points the market can realistically “give” on average on each timeframe.
These are points, not pips.
Our current goal, before moving on to longer-term forecasts, is to consistently show high accuracy on short-term moves. That’s why we’re working with fast data. There’s no need to wait days or weeks — we can see the results today. All trades were accurate, and the statistics have been recorded.
It’s important to note that our metric is quite different from what is usually found in typical EAs (at least we haven’t seen anything similar), and it scales well.
If we want to get more data from the market quickly, we physically cannot aim for a large number of pips. There simply aren’t that many pips available in 5–15 minutes. Of course, we could bring in an excavator, but a shovel is more than enough — to use a metaphor.

As a result, the AI is tuned for a small number of pips, and it does exactly what it was tuned for

However, we can clearly see that these structures are not perfect. Some are larger moves, some smaller. The EURJPY trade, for example, was placed during the Iran news period. Hypothetically, that structure held for some time and managed to deliver exactly what we asked from the AI.
It’s obvious that the news had a clear impact (the trade was entered right at the beginning of the news, which is visible on the M1 chart).

In short, the AI is doing exactly what we ask of it, and it’s doing it quite well.
Another successful day.

The real question is what we want from it. At this point, it’s no longer about the AI itself, but about the structure it caught and made a forecast on. What kind of n-dimensional structure is this? That’s something we need to analyze more carefully in context. In this sense, the AI is now acting as one more opinion alongside others, before we make a longer-term decision — the kind where you can actually extract more pips from the market.

Hey, thanks for the explanation. The reason I’m asking was I would like to see a balance profit versus loss ratio since I don’t see the stop loss being placed in the first place.

Thank you for the discussion — I’ll try to add some perspective based on our current observations.

In one of the screenshots (AI mode F9), we presented a histogram showing the distribution of trades by holding time.

Looking at it overall, the distribution appears to be significantly left-skewed:
the majority of trades are closed within a relatively short time frame — roughly in the 5–60 minute range. At the same time, there is a clearly decreasing pattern, with a strong concentration in the earliest interval (for example, trades under 15 minutes account for 557 observations versus 56 in the next interval), followed by a gradual decline.
At the same time, there is a smaller group of trades that extend beyond this range and remain open for significantly longer.
These cases are of particular interest to us.
From an observational standpoint, such trades may correspond to situations where:

1. market movement becomes less directional
2. short-term momentum decreases
3. or the volatility regime shifts

We do not treat these cases as “system errors”, but rather as areas requiring additional interpretation. Supporting this view is the shape of the histogram tail — it appears smooth and gradual, without sharp outliers, which may indicate a more natural underlying process.
At the current stage, in such situations we additionally evaluate context factors such as:

1. the structure of price movement
2. current volatility conditions
3. overall instrument behavior

Based on this, decisions regarding trade management are made.
It is important to note that there is no universal rule here — different situations may require different approaches, and the level of involvement in trade management can vary.

Different systems and trading algorithms (EAs) offer a wide range of solutions to this problem. We are developing and researching our own approaches, some of which are already showing interesting results and are discussed in a more closed setting.

Overall, stop-loss handling remains a complex and important area that requires continuous research.
In other words, this is one of the aspects where automated logic meets interpretation, and these boundary cases are currently of the greatest research interest to us.