When The Numbers Finally Matched

A tiny win

There was a day in the MyTradePilot build that did not feel like a breakthrough at first.

No big launch. No confetti. No sudden feeling that I knew what I was doing.

The scanner just returned candidates.

Actual candidates.

That sounds small, and it was small, but after what came before, it mattered.

Earlier in the build, I had pressed the scanner button and got exactly zero results. Nothing. Blank. The AI explained that the problem was in something called mockMarket.ts, and that a seeded random-number generator had a hash function returning a float where an integer was needed.

I am 50 years old and have no coding background, so I will be honest: that sentence might as well have been a weather report from Mars.

All I really knew was this: I clicked the button, and the thing I was building did not work.

That is a strange feeling when you are making software with AI. On one hand, it feels almost magical that you can describe what you want and watch screens appear. On the other hand, when something breaks, you are staring at words you do not understand, trying to decide whether to trust the fix.

Then there was the P&L scare.

At one point the app showed a $124,000 phantom gain. Phantom is the important word there. It was not real. It was a number on a screen pretending to be real, which is much worse than a button being ugly or a page loading slowly.

The whole point of this app is trust in numbers.

If a trading scanner gives bad numbers, that is not just annoying. It is dangerous. A crash is one thing. A wrong number that looks believable is another thing entirely.

So when the scanner finally returned candidates, I did not celebrate right away. I got suspicious.

I opened a spreadsheet.

This is probably the most “me” part of the whole story. I may not understand a broken seeded random-number generator, but I do understand checking math in rows and columns. So I took the numbers from the app and worked them out by hand in the spreadsheet.

Entry price. Target. Stop. Possible gain. Possible loss.

Nothing fancy. Just the basic profit and loss math that had to be right.

And this time, it checked out.

I remember sitting there with that quiet little feeling of relief. Not victory. Relief.

The kind where your shoulders drop a bit and you realize you had been holding your breath.

It was not the end of the problems. Not even close.

I was still obsessing over the logo like it was somehow the entire company. I had already paid for logo generation more than once, rejected the first one, asked for five more, picked a paper-airplane-and-candlestick idea, then changed my mind and uploaded my own image anyway.

I had also renamed the app from QuantisAI to MyTradePilot, which seemed simple until I noticed the site had three different names floating around. One page said TradePilot Academy. The homepage said MyTradePilot™. Other parts still had old branding energy hanging around like a bad haircut in an old photo.

And then there were the security warnings with names like SUPA_anon_security_definer_function_executable and search_path_mutable.

I still do not know how to say those out loud confidently. I had to let the AI explain and fix them, while quietly hoping I was not shipping a security hole with a pretty button on top.

But that day, the scanner worked in a way I could verify.

That mattered.

I had also recently agreed to my first recurring bill for the market data and infrastructure behind the scanner. A monthly cost made the project feel less like tinkering and more like a real commitment. So seeing actual candidates appear made that decision feel slightly less ridiculous.

Not completely unridiculous. Just slightly.

The fear is still there. Especially the fear that a real person could lose real money because of a bug I did not catch. That fear is different from worrying the app might crash. A crash is obvious. A bad calculation can wear a suit and look professional.

So I am trying to build this slowly. Check the math. Ask smaller questions. Make fewer giant leaps.

If I were talking to someone starting today, I would say: do not try to build the whole thing in one prompt. Take small honest steps.

And maybe pick the name first.

Also, stop redesigning the logo.

One small checked number can feel better than a hundred shiny screens.