Meet Billie

Start with what I did today

I built an AI trading partner called Billie, and I use it every day.

Today I turned off all eleven of its alerts.

These had been firing every day for months. Gamma structure shifts, walls moving, certain patterns appearing — all of it went to my screen. I switched the whole set off.

Because when I finally graded them properly, not one was worth keeping.

Opening an introduction to Billie this way probably looks odd. Introductions usually start with "here is how this tool helps me."

But this is precisely what Billie is for. So this is where I start.

1. What Billie is

In one line:

A system that records my judgments every day and later grades whether they were right.

Billie does not recommend securities. It does not produce buy or sell signals.

What it does is observe and record what today's market structure is, remember what I said yesterday, and check later whether that held.

The parts:

Part	What it does
Observation	Logs gamma structure and flow in the options market, minute by minute
Morning report	Before the open, lays out today's structure and its boundaries
Rule check	Flags an entry that violates a rule I set
Record	Saves my reasoning at the moment of entry, as written
Grading	Automatically evaluates that judgment once time has passed
Review	After the close, compares what I said against what happened

The last two are the point. Other tools observe and record. Grading and review are why Billie exists.

2. Why I built it

Market data is everywhere. Nobody records my judgment

Plenty of services sell options market data. Several show gamma structure, and most of them are built far better than mine.

None of them record this:

"Why did I decide that at 9am yesterday?" "Was it right?"

After three years of trading, I had the feeling that I was repeating the same mistakes without being able to say which mistakes, exactly.

A feeling is not evidence. So I started keeping records, and then made those records grade themselves. That is Billie.

The second reason — I don't trust my memory

Trading journals written by hand get edited without you noticing.

Winning days get detail. Losing days get a line. You write "I entered because of X" when X is often a reason reconstructed after you knew the outcome.

Billie saves the entire market state at the moment of entry. Whatever I remember later, the numbers from that moment are still sitting there.

I have checked this repeatedly. My memory and the record disagreed. Usually my memory was the wrong one.

3. How Billie actually helps

① It brings back rules I wrote and forgot

This happened today.

I was about to restart with a smaller account, and Billie warned me the risk rule didn't fit. I pushed back. And I was right — Billie had confused collateral with expected loss.

It conceded. But instead of stopping there, it continued:

"So the real question isn't the size of the risk. It's how many days of ammunition it represents."

And it pulled up a sentence I had written myself two months earlier:

"The account ceiling is a last line of defense, not a daily rule. The box empties in one day and there is nothing left to recover with."

I had forgotten that sentence. My past self argued against my present self. Billie only connected the two.

That is what working with AI actually looks like to me. Not an oracle handing over answers — something that remembers my own record better than I do and joins the pieces I dropped.

② It runs checks at a scale I can't

While grading alert accuracy today I saw something strange. Four signals had a 100% win rate. Fifty-five for fifty-five.

I was not pleased. I was scared. Nothing is right 55 times in a row in this market.

Digging in with Billie, we found a missing ticker filter in the scoring code. The S&P 500 and QQQ were sitting in the same column, so every down signal was being marked correct unconditionally.

I re-scored 329 records. The 100s collapsed to 7–36%.

I would not have found that alone. Tens of thousands of lines, and the grading logic is not somewhere I look daily.

But Billie didn't find it on its own either. It started with my question — "What exactly are these four 100% signals measuring?"

③ It found an error in my own book

In a book I published in 2023 I wrote this principle:

"Risk to target reward should be at least 1:3."

While working through risk math with Billie I realized: that is a standard for buying options, and it is structurally impossible when selling.

And selling is what I do every day. My volume-one principle and my volume-three strategy had been contradicting each other for three years.

The revised edition will carry this calculation as an appendix.

4. How Billie grows

Billie is not a learning model. It does not get smarter by itself.

Rules accumulate instead.

Whenever something goes wrong — a loss, a bad judgment, a bug in the tooling — I write down why. And Billie reads that document every time.

So Billie's documentation is mostly a list of failures.

Hit the day's target, got greedy, added more, lost badly → "Hitting the target is not a reason to exit."
Structure changed intraday and I kept pushing the morning read → "When the regime changes, change the strategy immediately."
Same day, same strategy, different placement — one made money, one lost → "Right direction, wrong location, still a loss."

There are dozens of these now. Billie reads all of them before each morning starts.

And I keep the rejected ones too. That may matter more.

A good number of my hypotheses failed validation — plausible on the surface, gone once the sample widened. Those stay on file together with why they were rejected. Otherwise I'll have the same idea again in six months.

5. What Billie can't do — this part matters more

Plainly:

① Billie can't call direction. That's why I turned off all eleven alerts today. The best one scored 37%. Predicting direction is not what this system is good at.

② What it is good at is boundaries. It can say "today likely stays between here and here." It cannot say "this goes up." I try to hold that line.

③ Billie gets things wrong. Today alone it was wrong on the risk math, and I corrected it by pushing back. A wrong objection brought the right question.

④ Billie does not make me money. It records whether I followed my rules. Following them is on me, and so is the outcome.

6. Why publish this

Writing and coding with AI is something everyone does now. That's a baseline, not an edge.

What I think is hard to copy is a different thing:

Working to disprove your own idea with data.

Not stopping at "this signal looks decent" but going on to "how would I confirm this is wrong?" — and when it comes back wrong, switching it off.

Turning off eleven alerts left the screen quiet. They had been part of every day for months.

But silence beats a wrong alert. Especially thinking about what would have happened had I traded a fake 100% signal.

What goes here next

I plan to keep recording the work with Billie on this site.

Rule log — did I keep the rules today, or break them. Rules, not returns. A losing day still counts
How a rule is born — hypothesis, test, accept or reject. The rejected ones stay too
The days Billie was wrong — probably the most interesting ones

I don't plan to write "how to make money." I don't know yet either.

What I'll write instead is "I tried this and here is what happened." The days it worked, and the days it didn't.

Billie is my personal trading record-and-validation tool. It does not provide investment advice or trading signals. This is a record of building and using that tool, not a recommendation to trade any security. Your decisions and their outcomes are your own.