Why I built this
I knew how to build a house. I had no idea how to build a trading system.
I was 44 when my first grandchild was born.
That grandchild lives in another state. When you make your living in construction, getting away takes more than clearing your calendar. Time off costs money, and the work still needs to get done.
I started thinking about how I wanted to spend the years ahead—and how much of them I wanted to spend missing things because I had to work.
That's what led me to the hardest job I've ever taken on.
An automated trading system.
When I started, I had never written a line of code. I didn't know what an API was. I had construction experience, a reason to keep going, and access to Claude, ChatGPT, and Codex.
I figured I could learn as I went.
I did. Just not in the order I expected.
Everybody was working. Nothing was finished.
At first, the speed was incredible. Describe something to an AI, and it would write the code. Ask for another feature, and it would build that too.
Then I started asking one AI to check another's work.
Something declared finished would turn out to have problems. Fixing those problems would produce another version. Another review would uncover something else.
Eventually, I had 364 copies of the project on my computer.
I had systems watching other systems, safeguards around safeguards, and reports explaining why I needed more of both.
What I didn't have was a working trading program.
The market recordings were another problem. During the first month, ten days of data were lost or corrupted. Streams stalled. Connections timed out. Windows refused to start scheduled tasks. One night, the watchdogs meant to protect the recordings nearly cost me an entire day of them.
I understood the situation better when I stopped looking at it as a software project.
On a construction site, a crew can stay busy all day without getting the wall built.
Activity doesn't mean progress. And sooner or later, the owner has to walk onto the job and ask: What actually works?
I started running it like a job site.
I put the priorities on one page:
Build a working trading program.
Make every step produce something observable: recorded data, a trading decision, or an order at the broker.
Add a system only when a measured problem requires it.
Call something PASS only when the evidence supports it.
I changed the division of work, too. One AI builds. Another inspects the actual files and checks the result. I make the decisions and keep the project pointed at the goal.
That didn't eliminate mistakes. It gave me a way to catch them and finish the work.
It also meant accepting answers I didn't like.
Seven tests of ideas based on my own trading came back without evidence strong enough to move forward.
Seven times, the answer was no.
I wanted those ideas to work. But there wasn't much point demanding honest results from the AIs if I was going to ignore the results myself.
A monthly momentum strategy passed its five defined tests and became the Quant engine's baseline. That earned it a place in paper trading. Whether it holds up over time is still something it has to demonstrate.
We rebuilt. Now the results go public.
In October, my AI crew and I set aside the tangled build and started clean under those rules.
Within weeks, the system was recording market activity and placing paper trades automatically.
Today, Deather Vector has three engines, each assigned its own Alpaca paper account:
Scalper trades short-term stock moves and closes its positions before the session ends.
Quant manages a monthly momentum portfolio.
Longterm is designed for quarterly investing in an index fund and companies ranked by quality and value. Its first purchases are scheduled for December 1.
No human selects the individual trades.
Beginning November 2, 2026, the system publishes its trading records every evening after the market closes at deathervector.com: trades, holdings, and performance alongside the S&P 500.
The daily updates also go into a public GitHub repository so readers can inspect what changed over time. Earlier test trades are identified separately from the official paper-trading record.
The losses belong there just as much as the wins.
These are simulated accounts using real market prices. Paper results don't establish what would happen with real money. I'm not offering financial advice or selling a trading service.
I'm opening the books on something I built.
I don't know yet whether this project will give me the freedom I hoped for. That answer will take time, and the record needs to earn it.
But I know why I started: I want more chances to visit my grandchild without having to weigh every day away against the next job.
For now, I'm still working construction, still learning, and still holding the system to the same standard I'd expect from any crew.
The wall has to get built.
And if I'm going to tell people it's standing, they ought to be able to come look.
Follow the daily trading record at deathervector.com. To receive updates, subscribe on the site or at deathervector.substack.com. Both join the same mailing list. I'll write when there's something worth reading.
Update, November 4, 2026: The Deather Vector Substack page is temporarily unavailable. To get updates, sign up on deathervector.com.