I built a feature to verify trading logic from YouTube videos

YouTube is full of “you can win with this method!” videos. They’re tempting, but verifying each one by hand is a lot of work. So I built a tool that extracts the trading rules from a video and feeds them straight into the testing framework.

How it works

Roughly, the flow is:

1. Give it a YouTube video URL
2. Have an AI (Gemini) watch the video and extract the mechanizable rules (entry conditions, exit, stop-loss) as JSON
3. Run those rules through the framework — all the way through forward testing and the M1 intraday check

The key is to extract only the parts a program can reproduce, not the “vibe” of the video. Vague phrases like “around an obvious high” get turned into concrete conditions where possible (and where they can’t be, that’s exactly the part that depends on human discretion).

What I found

I ran several famous video strategies through this. Honestly? Most of them “vanish once mechanized.”

The common patterns:

• EMA cross + pullback + “wave 3” systems: look good on the one timeframe they were found on, but fall apart on other timeframes (i.e., they fit that timeframe’s noise)
• Counter-trend / false-breakout fades: the edge disappears in forward testing
• The real earner is discretion: the “reason it wins” depends on drawing lines or counting waves — human judgment that doesn’t survive being reduced to rules

This doesn’t mean the videos are lying. Often the creator’s own discretionary skill is doing the work, and that simply can’t be transferred to an EA.

Why it’s useful

With this tool, every time I spot an interesting video I can quickly run “extract → test → result.” Before spending time or money on a dubious holy grail, I can honestly check it on my own framework — and that’s the biggest win.

Verification is done on the rules disclosed in public videos. Paid course material is out of scope.