Increase the decision of the high-win-rate EA "Mikazuki" to discretion. AI-powered indicator "Mikazuki Eye"
Even among candidates that did not reach 100, there are chances that can be utilized through discretion. I did not want to let that potential go to waste.
Hello. I’m yoru, the developer of Mikazuki. I have developed the “Mikazuki AI,” which turns the判定 logic of the AI-powered EA "Mikazuki USDJPY" into a form usable for discretionary trading.
The reason for creating it is clear. I want to make better use of the information Mikazuki extracts from the market. Even if it does not meet the criteria for an order placed by the EA, there are people who can layer their experience and analysis onto those buy/sell candidates. To those people, I wanted to deliver this decision material.
↓ The indicator introduced this time, “Mikazuki AI” is here↓
https://www.gogojungle.co.jp/tools/indicators/86367
↓ AI-powered EA “Mikazuki USDJPY” is here ↓
https://www.gogojungle.co.jp/systemtrade/fx/79530
It would be a waste to end these candidates without using them.
First, please look at the actual chart. In the image below, the “Sell 78” and “Buy 36” that have not reached the confirmation of 100 are displayed.
After “Sell 78” there is a decline. After “Buy 36,” there is a brief drop followed by a rebound. Neither is a definitive 100% signal. Still, these are moments where price movements followed the displayed buy/sell directions.
Seeing such candidates, I think: it would be a great loss to leave Mikazuki’s拾い上げた information dormant merely because the EA’s entry conditions were not met!
Of course, sometimes, as with the Buy 36 example, prices move in the opposite direction after the candidate. This creates room to compare with the price range I am watching and the reversal pattern. I believe that one extra step is where a discretionary trader’s experience can add value.
There is a reason why the EA seeks “100.”
In the Mikazuki EA, the AI conditions corresponding to the overall AI成立度100 in the AI, and also trading conditions such as time of day, news, and spreads must be satisfied for the scene to be considered.
The EA mechanically repeats fixed rules. A trader cannot decide in real time, “Let me wait until this horizontal line reverses.” Therefore, it is necessary to verify performance by fixed criteria and strictly screen at the entry to protect the operating policy.
Here, 100 indicates that the necessary AI conditions have been met. It does not mean a 100% win rate. I value the stringent criteria that should be protected in automated trading. At the same time, I believe there is value in the information just before those criteria for discretionary use.
Overlay your rationale onto the AI-found candidates.
With discretion, you can layer higher-timeframe trends, lines you drew yourself, recent price movements, and the timing to wait for an entry. Mikazuki’s buy/sell candidates align with the conditions you specialize in. To pick the moment, I made information below 100 useful.
My goal was to have Mikazuki’s judgments overlap with the user’s experience. You can wait for carefully selected confirmed signals, or you can consider candidates where your own reasons align. Decide whether to trade, wait a little longer, or pass. The choice is left to the trader’s discretion.
Even candidates under 100 can be effectively used with discretion. I see value in shaping that potential as an indicator. I want to also utilize the AI’s information in conjunction with the decision-making ability of the user. That is why I created the AI.
I focused on making it possible to see the judgment partway through.
If you hand candidates to discretion, numbers alone aren’t enough. It must be clear what conditions are met and what hasn’t yet reached the target. Only then can I believe you can combine it with your own analysis.
On this screen, RSI meets the sell-candidate conditions, but the Candidate-fit AI is 80%, the win-condition AI is 68%, and two items are not reached. Because the missing conditions are displayed, you can see why to wait and compare with other grounds. The overall AI成立度 represents the lowest attainment among the required AI conditions.
Internally, there are seven series of entry judgments and a total of 210 rules for entry, plus 80 initial TP/SL mode choices, totaling 290 trained decision trees. By combining RSI across multiple periods and its changes, the bodies of confirmed candlesticks, higher-timeframe RSI, and price positions relative to the past 200 candlesticks, the candidate is evaluated by model groups according to direction and state.
I summarized these complex judgments into a dashboard on the chart and with tabs that can be switched. You can set the minimum number of candidates to display yourself, and hover over past marks to review details. I wanted it to be more than just looking at numbers; to tie into your trading decisions.
Open the judgment of high-win-rate EAs to discretionary use.
The real account listed on GoGoJungle for Mikazuki USDJPY as of September 15, 2026 shows a win rate of 83.02% (44 wins / 53 trades), PF1.80, and profit of 109,653 yen.
Mikazuki AI is designed so that you can combine this EA’s v9 judgment logic with your own analysis. For Mikazuki users, to confirm candidates and成立条件. For those who do not own the EA, to add AI decision material to your discretion. The AI can be used on its own.
I want to use Mikazuki’s judgments more fully.
Smartphone notifications for confirmed signals, initial TP/SL guide lines for relevant new manual orders, and a Japanese manual with actual screen included are prepared. In the Windows MT4 version for USD/JPY on the 5-minute chart, we’ve organized it from finding candidates to considering trades for daily use. The guide lines are displayed as reference prices, and do not set TP/SL or automatic settlement on orders.
How you read the candidates Mikazuki finds, how you layer it with your own strengths. That is why I created a product that makes even information before成立 visible.
Details on AI judgment design, backtests, and publication results: Mikazuki USDJPY ↗
↓ The indicator introduced this time, “Mikazuki AI” is here↓