I tested 50 trading bots with Claude. One completely dominated. A practical breakdown on building profitable trading strategies with AI (full guide):

How did it handle slippage once live?

the best of 50 backtests measures how many you ran, not what the strategy knows

Those days of crypto banter, gaming coins and fan tokens. Now we evolved into using AI

testing 50 variations until one overfits the backtest isn't a strategy it's just p-hacking with an llm wrapper

You only farming engagement with your videos nothing else . Nobody use your system to build anything. One day you said Claude is the best next day you said Astra ,Geok bot . Etc

@milesdeutscher so true, figuring out profitability balance was a game-changer for me with AI. Sometimes it's about knowing when to pivot, not just stacking features.

I am so happy that I am unemployed so Claude cant take my job, only save me time trading

Agreed

man has been dominating automated trading

dominated on paper, or survived fees and out-of-sample testing?

You make me feel so behind

Howch much is Profit Factor? Ssharpe Ratio Backtest?

Testing the options is the fastest way to find what actually works.

Have you tried any of them with real money?

Hey Miles, would love to have a chat 🙏

which one won? Gonna need the backtest stats before I believe any AI hype.

Of fucking course!

The missing benchmark is not win rate—it’s how often the strategy survives regime change. A bot that dominates one slice can just be overfit with better marketing. Log drawdown, slippage, and the no-trade decisions too.

Engagement farming FTW😂

now run the winner live with fees and slippage, that's the real test

If you don’t use purged kfold on the backtest while also accounting for the number of trials you are just overfitting on the sample data.

Another slopfluencer. What another waste of resources.

the only spot with a big red "?" is the live candle, everything green is already behind the price

did it dominate live markets, or just the test set?

The one that dominates 50 tests is usually the one best fitted to the sample. The number worth publishing is live PnL after fees and slippage through a week like this, where total mcap dropped 4.8% in a day.

survivorship bias in backtesting is key here. curious about the walk-forward analysis.

50 bot mà chỉ có 1 em vượt trội thì đúng là đáng nể thật 😄

did the winner still dominate after fees, slippage, and a real out-of-sample test tho

Passa pra nós.

What criteria did you use to determine which of the fifty trading bots performed the best?

Curious whether the winning strategy leaned more toward trend following or mean reversion.

Did the winning bot ditch all the overfit indicator combinations the other 49 were using, or did it actually surface a signal no one else thought to test for?