Mind your Markets

Most retail traders quit here... don’t give up!!! Level up your skills and knowledge.

I started this to prove I could build something that worked. What I actually built was something that tells me the truth even when I'd rather it didn't. That turned out to be the more valuable machine.

If you take one thing from all of this, take the separation between being right and making money. They're different problems with different solutions, and confusing them is the most expensive mistake available to a retail trader.

That's the arc. A system that's right most of the time, loses slightly, fails closed by design, and shows its work. The next chapter is fixing the skew, and I'll post that one the same way.

The number I'm least proud of is the one I've quoted most in this series. Down $6.60 on a 61.3% win rate. I keep saying it because a system that only reports its good weeks isn't a system. It's marketing.

Every blown-up trading product has the same shape. Nobody could see the code. The risk checks were optional and nobody knew. The backtest was chosen after the fact. Closed source isn't just opacity to customers. It's opacity to the person who wrote it.

Went back through what a closed system would have let me get away with. Cherry-picked windows. Quiet formula changes. A win rate quoted without the P&L that contradicts it.

I also published it because secrecy is a very effective place to hide from yourself. It's much harder to keep a comfortable story about your performance when the formula is public and the settlements are listed.

In most markets, publishing your edge destroys it. Weather is different. If you run the same forecast blend against the same contracts, the atmosphere doesn't become less predictable because two of us are watching. The edge is in the physics, not in secrecy.

Publishing week. The whole system is open source. Every risk gate, every sizing rule, every place the accounting could go wrong is readable by anyone who wants to check my work.

I had a version that reported a healthy profit. It was a formula error that made every two-sided position look like a win. Finding it felt like the floor dropping out. Every number I'd believed had to be re-earned.

I compute realized P&L from settled contracts alone, with one formula, in one place. Not from fills, not from unrealized marks, not from any field the API populates inconsistently. The hardest discipline in this system is refusing to derive that number twice.

Rebuilt my P&L calculation from settlements rather than fills. Fills tell you what you did. Settlements tell you what it was worth. Only one of those is the truth.

My worst accounting bug produced a P&L that was wrong in my favor. I believed it longer than I should have, because I wanted to. Being wrong in a flattering direction is much harder to catch than being wrong in a painful one.

The expensive hallucinations are never the obvious ones. They're the plausible field name, the slightly wrong sign convention, the fee applied at the wrong stage. Each one produces numbers that look reasonable and are quietly false.

Auditing generated code against the actual broker API. This is the unglamorous half of building with AI. Every assumption has to be checked against a real endpoint response.

I lost weeks to code I didn't read closely enough because it looked finished. The failure wasn't the tool. It was that I'd outsourced my judgment to something very good at sounding certain.

The pitch is that you describe a trading bot and get one. What you get is code that looks correct, imports libraries that exist, and contains reasoning errors exactly where you're least equipped to check. Plausible and correct are not the same property.

Back to the system. The part nobody puts in the build log: what it actually cost to assemble this with AI assistance, measured in time rather than dollars.

I couldn't resolve it and I'm not going to pretend I did. What I took away was narrower and more useful: I only want to trade contracts whose settlement source has no opinion about who wins.

Both sides of that argument have financial exposure to its outcome. That's the actual lesson for a trader. When every commentator on a market is also positioned in it, commentary isn't information. It's inventory being talked up.