Top Secret Project - A World with 0.01 pips of Error Margin

[Important]

This project is currently in the development and verification stage of building and validating a general AI and its mechanism on the n_k_i___ side, and it is not guaranteed to be realized.

However, even if it is not realized, verification has progressed to a stage where there is some possibility of offering in some form such as a Python × MT5 integrated indicator, building a separate server, or limiting to specific currency pairs.

There may also be operational challenges beyond the logic, such as handling the learning model, scale, and operation methods.

We are at a stage where we would like to offer it as an official collaboration product between Gogojan × Onisoku AI, as much as possible.

Since this is about the future, it is difficult to say it is “alchemy,” but in terms of what can be realized by individuals who have no skills, assets, or facilities, it might not be an overstatement to say it is close.

※ The data introduced below is verification data based on Onisoku AI. In other words, it is a story of “somehow it’s amazing what Onisoku AI can do.”

【Main topic of the article is below】

Installing Onisoku AI (AI learning model) on Gogojan servers and performing market learning and stable distribution on servers where it is practically impossible for individuals, not only on MT4,will enable notifying beginners to advanced users that at certain points you can gain up to a certain number of pips, with a predicted accuracy of a certain percentis ongoing.

Currently, the indicator that displays signals on MT4 isindeed achieving good results even with a non-discretionary EA.

For traders at intermediate level or higher, it is not difficult to perform basic discretionary trading by comparing the signals with their own methods to enter and execute efficient exits.

However, among beginners to early-intermediate traders who have been trading for a few years,they often hesitate whether to trust the signal or not, and they quickly bail out at a small drawdown, leading to instability in tradinga lot.

Therefore, we aim to elevate OniSoku AI’s microAI to a Neo-Onisoku AI form and make it morebeginner-friendly by delivering in real time “win rate,” “expected pips,” and “expected drawdown pips”in real timeas a project.

And when we recently reported the progress of this project,we received a considerable response, so we would like to introduce some of the technologies and verification aspects that are usually not touched upon.

First, this graph shows the error and correlation between predicted pips and actual progress.

In conclusion,it is “highly correlated”to some extent.

Aside from indicators or statements that disregard technical signals altogether,the average error is around 15 pips, and 80% of cases fall within 20 pips of error.

In broad terms,if the AI predicts a signal will extend by 80 pips, it can be said to extend by 60–100 pips with an 80% win rate

Of course, since the future is uncertain, misses can occur, but this includes the pleasant overestimation as well.

The next graph ranks predicted pips and checks whether there is no error within each range.

For example, a model that predicts 40 pips correctly but 100 pips incorrectly is not useful.

As you can see, OniSoku AI demonstrates high-precision predictions across each predicted range.

Error 0.001 pips

Of course, there are misses.

On the other hand,there are multiple cases where the minimum error is 0.001 pips, i.e., an error of 0.00001 yen.

By the way, in the past 3–4 years of verification data (all training data), there were 4,288 signals,within 1 pip of error there were 500 cases, and 1180 cases had an error of 20 pips or more.Since the error includes “pleasant miscalculation,” half of the 600 cases were “greater than expected”. On the other hand, 600 cases did not extend as much as expected.

Even if we can predict with perfect accuracy, exits should be done progressively, using built-in stops, trailing stops, and partial exits as the basics.

Now, while Onisoku AI’s main features are the signal function and market syncing, in short, it issues signals where such growth is expected.

However, on MT4, it is almost impossible to display stand-alone how far a signal will extend and how confident it is.

Therefore,if this project is realized, a wide range of people will be able to set clear criteria such as “where to place a limit order,” “where to cut losses,” and “whether to discard if win rate falls below a certain percentage.”

By leveraging AI, we are entering an era where anyone can draw like an artist, take photos like a photographer, and write novels like a novelist. Howeverwith generative AI, trading like a top investor is not yet possible.

We hope OniSoku AI will help you trade like a pro trader.