ai trading / algorithmic trading / backtesting / live trading / public portfolio challenge
Every Rule My AI Trading Bot Follows Is Public. So Is the 12% Crash.
In February 2026 I put $25,000 of my own money into a live Public brokerage account, wired it to an AI trading system on NexusTrade, and made the whole book public: every position, every fill, every model test, every bug and every failure. This video walks through what actually happened, in order.
The system does not predict prices. It writes candidate strategies, backtests them across a corpus of prior runs, and then tries to break them out of sample. One variant it found returned +6,130% in a 2023-2026 backtest and was rejected on a -65.4% drawdown. The one deployed instead returned +59.1% over 2025 against the S&P 500's +23.6% in the same window, with a -32.7% worst drawdown of its own. Those are backtests, not results.
The live account went on nine days later and was up 16.6%. It then gave almost all of it back, falling from $29,171 to $25,570 over thirteen days in May. At the peak of that run an audit found a real configuration error in my own exit logic, where NVDA had been left out of the underlyings filter on a close rule, so a take-profit that should have fired never did. I fixed it, wrote up the mistake including the manual sale I should not have made, and the account has gained about 19% from that low.
In August I replaced all twenty-two strategies with a raw-return design, and recorded in the decision memo, before deploying it, that it was not a lower-risk upgrade. Three days later the book hit an all-time high of $32,305 and eleven days after that it was down about 11%. Every figure here is checkable: the account is shared, the rules are not obfuscated, and the research repo carries the losses alongside the wins.
Transcript
0:00This guy is up more than 20% using an AI that trades stocks for him, and his entire method is published online for free.
0:08Most people who try this just open a chat and ask it what a stock will do tomorrow.
0:12He built the opposite of that. His system writes the strategies, backtests them, and then tries to break them on years of data it has never seen.
0:20It found one that returned 6,000%, and he threw it out, because it drew down 65%.
0:25The one he kept made 59% in 2025, against the market's 23%.
0:31On May 5th he turned it on with real money, and nine days later he was up 16%.
0:37Then it gave almost all of it back, falling from $29,000 to $25,500 in 13 days.
0:43He had already found the problem at the peak, an exit rule that was silently skipping NVDA, and he published that mistake while the account was still falling. It has gained 19% since the low.
0:54Then in August he replaced all 22 strategies, and before he switched the new set on, he wrote in public that this one would drop harder.
1:02Three days later it hit an all-time high, and 11 days after that it was down 11%, exactly as he predicted.
1:10He is still up more than 20% on real money, and you can check every number I just said, because the account, the rules, and the losses are all on his GitHub.
1:22Comment PROOF and I will send you the link.
Join the conversation
Loading conversation…