[Verification Record] The complete log of a 40-something office worker who can't read code partnering with AI (Claude) to complete a "moving self-made EA"

Chapter 1: Introduction —

The Slow Life I Admire and the Trading Life Hit by Walls

I had been working as a company employee for many years, but I suddenly began to think, “I want to leave the忙しい city life and spend more time on the things I love and my hobbies, living a slow life.”

To achieve that, I needed pillars of asset formation that are not bound by time or place. I had long been interested in FX and automated trading (EA), but I had a fatal flaw: “I have absolutely no programming skills (MQL code).”

Because I can’t write code myself, trading was entirely manual or with limit orders. However, I’m busy with my full-time job and housework, so I couldn’t sit glued to charts. “I was so close to hitting the limit order…,” “If I had been watching the chart, it would have been a chance…” I repeated such frustrating thoughts many times, and as the results didn’t go as I hoped, I was halfway giving up.


The initial failure with “ChatGPT”

Then the AI boom arrived. “If I can’t code, maybe I can get AI to write the code?”

Clutching at straws, I first used ChatGPT to construct trading rules and have it output EA code. I set a perfect rule (supposedly) to minimize risk to the extreme and profit safely through backtesting, and I投入 a demo operation.

…But after a month, there were zero entries.

In my attempt to eliminate risk too much, the entry conditions became too strict, turning the demo account into a statue that barely moves. I tried several times by changing prompts, but I couldn’t get it to behave as I hoped, feeling frustrated.

Meeting “Claude” that Changed My Destiny

“Is it really impossible to create an EA with zero knowledge…?”

Just when I was about to give up, I saw talk in the engineering and AI communities about “Claude from Anthropic” having astounding accuracy in code writing.

“If Claude, who excels at code generation, can shape my ideal EA, perhaps it’ll become a reality.”

Still skeptical, I decided to try one more time.

Chapter 2: Interaction with AI (Claude) and Development that Wasn’t Straightforward

Learning from ChatGPT’s painful failure of an “EA that never enters”, I decided to rethink how I work with Claude from the ground up.

For me, who can’t read code, AI isn’t just a “code output tool” but a partner (a skilled engineer) with whom I can flesh out the specifications together.

The EA I developed this time, code-named “ORB_RR_ (tentative),” was born through the following gritty trials and errors.

1. Break down “what I want to do” to the limit using natural language
Because I have no knowledge of MQL5, I described concrete trading logic in everyday Japanese conversation, such as “based on Tokyo time high/low, ~” and “risk-reward is ~” in a straightforward way. Claude translated my vague words into a surprisingly elegant logical structure.

2. Debugging without panicking when errors occur
Of course, it wasn’t perfect on the first try. There were compile errors in MT5 and backtests showing unintended behavior (stops not working, orders at unexpected times).

If I had done this before, I would have given up, but I copied the error messages and the odd behaviors as-is and fed them to Claude. “This error occurred—what caused it?” “It’s entering outside the specified time—what should I fix?” Claude didn’t frown (laugh), and quickly suggested the cause and the fix code.

Chapter 3: Finally, a “Moving” EA Is Born

After many rounds of backtesting and adjustments, the birth of “ORB_RR_Portfolio” finally arrived.

Unlike the previous failure (a one-month-zero-entry statue EA), it functioned properly in backtests, with a balance of risk management and entry frequency, and a “living code” that moved as if it had a life. I couldn’t help but do a fist pump in front of the screen.

A forty-something corporate employee with no programming knowledge obtained my own original EA by properly treating AI as a partner.

Currently, I have deployed this “ORB_RR_Portfolio” to a demo account and am conducting forward testing (live market validation) in real market conditions.

Chapter 4: Conclusion (Next Time Preview)

Next time, I will comprehensively explain the conceptual framework, broad logic, and trading conditions of the “ORB_RR_Portfolio,” which even I, who can’t write code, managed to shape.
In the future, I plan to deliver real-time data on monthly demo operations and real operations with no fudging.

I hope this will help those who feel they have no time or knowledge. Please kindly follow the lab!