Yinmusubi AI Development Story Episode 4 — What to entrust to AI, what not to entrust (the "Enganch relation" authority)
Last time, we talked about protecting yourself with “stopping decisions” and “available capital” instead of a loss-cutting mechanism.This time, it’s about the AI that handles the “stopping decision.” But I did not entrust the AI with everything. On the contrary, I narrowed its authority to the limit. Why choose a “small-authority AI” rather than a “smart AI”? This is about drawing that line.
The only authority given to Enmusubi AI is“Stop/pause new purchases now”. That’s all.
When the market is volatile, moves sharply, or spreads widen, sensing these “danger signals,” the AI quietly stops new entries. The aggressive decision (where to buy, which asset) is determined by logic, and the AI only holds the gate of“whether to act now or wait”. In other words, it does not interfere with the attack. It is the “gatekeeper at the entry.”
The important thing is here. There are three authorities I did not hand over to AI.
- ① Do not cut losses on its ownAs described earlier, with a mean-reversion approach, you miss the recovery just before the drawdown ends. So I did not give AI the authority to cut losses.
- ② Do not change direction on its ownThis EA is buy-only at its core. I did not allow discretionary switching to selling based on AI’s intuition.
- ③ Do not halt ongoing recoveryThe riskiest thing is AI stopping further buys during a recovery. If it does, it pulls the rug right before it returns.
In short, AI’s authority is“to stop/wait for new buys from now on”.Do not touch any ongoing recovery processes at all. Not mixing this up is the key of this design.
Giving AI too much power can make it act in ways you didn’t anticipate, potentially causing damage. Intelligence can become a pathway to its own runaway.
If you limit its authority to “stop/wait,” even if AI makes a wrong call, the worst it can do is“do nothing (pass)”. It cannot take unnecessary actions that would wreck positions. This is the idea offail-safety. What I sought was not cleverness but a clear line between what could be done and what could not be done.
Note: The role of AI is not to attack, but to halt new buys when things look dangerous.
When AI decides to “wait,”Why waitis displayed on the chart panel in Japanese. I named this place “Enmusubi-do.” It shows reasons such as “currently postponing” or “no new purchases for now.”
Not just numbers, but the seasonal keywords of the night cherry blossoms (Hanabiyori・Hanakumori・Han hies) convey market temperature at a glance. Because the reasons are visible, even during drawdown you can feel, “Ah, it’s Hanabie (late spring cold) now.”Make judgments visible and plain to see— I will write more about this presentation method in the next piece.
AI’s job is not to attack.To stop new buys when it’s dangerousThat’s why I narrowed its authority. Not a smart AI, but a “small-authority AI” with clearly defined capabilities. Therefore, the worry of “I don’t know what it will do next” is minimized.
However, to be clear, repeating: even if AI stops new entries,existing drawdown is not eliminated. It is weak to a strong appreciation of the yen, and worst-case failure could occur. AI is not a万能 safety device; it is merely the “gatekeeper at the入口.” Please view it with sufficient available funds as a premise.
※This article is for information purposes and not an investment solicitation. The performance shown is historical and does not guarantee future profits. FX/CFD trading involves risks. Please make investment decisions at your own responsibility.