I Built an AI Trading Bot in Claude: Can I Make It Better? Part 2

I gave my Claude AI gold trading bot one prompt with every upgrade I could think of, then tested each one alone against the original backtest. Only ONE upgrade earned its place — and it cut the max drawdown from 21% to 12% without losing profit.

This is Part 2 of Vibe Coding MT5 Robots: an XAUUSD (gold) Expert Advisor built in Claude Opus 5.5 with zero hand-written code. In Part 1 it made +58.7%. With 3 more weeks of data it’s now up +103% (PF 1.97) — but a 21% drawdown was leaking money. Today we try to beat it.

📥 DOWNLOAD (ZIP): all prompts used in this video + the full EA source code (v3.1 & v4.0)
👉 https://ultimatetrading.tools/product/utt-ssl-fisher-matrix-ea-mt5/
▶️ Part 1:- https://youtu.be/Yi9ICWfmhgU?si=xMRY9ehvEw9ro3t5

What I tested (one switch at a time, 1 Aug – 7 Nov backtest):
✅ Friday cutoff (no trades after 12: 00) — DD 21% → 12%, PF 2.09 → KEPT (v4.0)
🟡 Anti-streak 50% — smoothest ride, highest PF (2.15), but −12% profit → reserve
➖ Sessions, News filter, End of day — never fired (retest coming)
➖ 1% risk per signal — sizing only, not an edge
❌ Runners with no take profit — worse
❌ Combining the two winners — they overlap, don’t stack

Join Telegram Channel- https://t.me/Ultimate_Trading_Tools
Channel 2- https://t.me/ShepherdFxTradingTools

⏱️ CHAPTERS
0: 00 The gold robot that made +103%
0: 40 The money leak: 21% drawdown
1: 00 My setup: Claude Opus 5.5
1: 16 How v3.0 trades (SSL Matrix + Fisher)
1: 35 What the backtest report tells us to fix
2: 06 One prompt, every upgrade as a switch
2: 34 Tests 1–3: Sessions, News, End of Day
3: 15 Test 4: Friday cutoff
4: 07 Test 5: 1% risk per signal
4: 36 Test 6: Anti-streak
5: 08 Test 7: Runners with no TP
5: 36 Combining the two winners
6: 17 Results scorecard
6: 25 The verdict: v4.0
7: 09 The out-of-sample test most backtests skip
7: 25 The method + downloads

Want Part 3 — live testing on a real account? Tell me in the comments.

⚠️ Disclaimer: Educational content, not financial advice. Backtest results do not guarantee future performance. Always test on a demo account first.

#ClaudeAI #TradingBot #MT5