You Built a Claude Bot. It Backtested Great. Then It Lost Your Money
Your Claude-generated trading bot passes 5 years of backtesting without a loss. The numbers look perfect. You deploy it live on Monday morning with $3,000.
By Friday, you're down $1,100.
This happens to traders all the time. Not because Claude can't write code—Claude absolutely can. The problem is that Claude writes strategy code. It doesn't write systems code. It builds what works on historical data. Professional developers build what survives in production.
Here's the gap.
The Backtest Trap
Backtesting is a lie machine. It lets you cherry-pick conditions, avoid slippage charges, and skip the friction of real trading. It tells you what worked in the past when what you need is what will work now—under real broker conditions, real spreads, real requotes.
Claude generates code that backtests perfectly. Professional developers don't backtest-and-pray. They:
- Backtest on clean data (not curve-fitted to the last 5 years)
- Forward-test in a live demo environment (real broker, real slippage, real lag)
- Start live with micro positions (1% risk, not 10%)
- Monitor live performance for 30 days
- Adjust parameters based on actual fills, not assumed fills
Here's what doesn't show up in your Claude backtest:
- Slippage on limit orders during news releases (you lose 5-12 pips instantly)
- Broker requotes that cost you 50+ pips mid-trade
- Liquidity gaps on lower-volume pairs at 3am London time
- Your strategy's sensitivity to volatility regimes it has never seen
- Partial fills that execute your 1-lot order as 0.8 + 0.2 separately (mess with your logic)
A professional EA tested live in demo for 2 weeks catches these. A Claude script deployed immediately gets blindsided by them.
Risk Management (What Claude Skips)
Claude will write you position sizing code. But it won't write the guardrails that separate "trader" from "one-week martyr."
Professional EAs include:
- Equity stop-loss — if your account drops 10%, stop all trading automatically
- Daily loss limits — if you hit -$500 today, stop trading
- Drawdown recovery — scale back position size if underwater
- Max spread filters — don't trade if spread exceeds 3 pips
- News event blockers — skip trading 2 minutes before economic releases
- Broker disconnect handling — reconnect and verify fill status automatically
Claude might write 1 or 2 of these. A production EA includes all of them. Here's the thing: a Claude bot with no equity stop will blow your account eventually. It just needs one bad market day. A professional EA is designed to lose cleanly, recover, and scale back. That's the difference between hobby code and professional systems.
Broker Integration (Slippage, Requotes, Disconnects)
You built your Claude bot against EURUSD. It trades USDJPY too. At 2am London time, the spread widens from 0.8 pips to 12+ pips. Your stop-loss gets rejected. Your position swings $500 against you in seconds.
Claude doesn't know:
- Your broker's slippage patterns by pair and time of day
- Which pairs trade liquid vs. thin at different times
- Broker-specific requote behavior (some requote on news, some don't)
- How your broker handles partial fills
- The difference between hedge mode and netting mode for your stops
Professional developers have tested this across 20+ brokers. They know that Interactive Brokers (IBKR) has different fill quality than TD Ameritrade, which has different infrastructure than smaller ECNs. They build with this knowledge baked in. A professional EA tested on your specific broker is not over-engineered—it's the minimum viable product for not blowing up.
US Regulatory Reality (FINRA, CFTC, NFA)
Retail traders in the US are bound by specific rules that Claude's general-purpose bot logic doesn't account for. Here's what matters:
1. The PDT Rule (Pattern Day Trader) — If your account is under $25,000, you can only day-trade 3 times per week. A Claude bot that day-trades on a $5k account will violate this and get you flagged. FINRA publishes the exact rules here. A professional EA knows this limit and builds around it.
2. Margin Requirements — FINRA margin rules vary by instrument. Your Claude bot might use 2:1 leverage on US indices, but that's the legal max. Anything higher on a US-regulated broker auto-liquidates your position. Professional developers know the exact limits.
3. Futures vs. Forex vs. Stocks — Different instruments, different regulators (CFTC for futures, NFA for forex, SEC for stocks). Different rules per category. Claude writes the same logic for all three. Professionals know the nuance.
FAQ: Is it legal to use AI to build trading bots in the US?Yes. Using Claude to write trading bot code is legal. Using that code to trade is legal too—as long as the bot complies with FINRA, CFTC, and NFA rules. The legal trap isn't "AI bots are illegal." It's "non-compliant bots are illegal." A Claude bot that breaks the PDT rule will get your account flagged. That's not a problem with AI—it's a problem with the code not knowing US regulations. Professional developers at Alorny build with these rules baked in by default.
The Production Layer (Monitoring, Logging, Alerts)
Your Claude bot runs. You go to sleep. You hope it works. At 8am, you check. Your position is half the size you expected. What happened? Did the broker reject an order? Did your internet drop? Did the bot crash? You have no idea.
Professional EAs include:
- Trade logging (every entry, exit, rejection is timestamped)
- Email alerts on critical events (position opened, position closed, error encountered)
- Equity tracking (daily P&L, win rate, average trade, drawdown %)
- Performance dashboard (optional, but we build them)
- Error recovery (if the bot crashes, it reopens positions from the last recorded state)
- 24/5 monitoring (logs tell you exactly what happened while you slept)
Claude generates code that trades. It doesn't generate the visibility layer that lets you trust your automation. A trader without visibility is just a sleepwalker. Professional builders give you the dashboard.
What A Professional MT5 EA Actually Includes
Here's what you get when you hire Alorny to build a $300–$500 Expert Advisor:
- Strategy logic (what Claude can do)
- Risk management guardrails (equity stops, drawdown limits, position scaling)
- Broker integration (slippage handling, requote logic, disconnect recovery)
- US regulatory compliance (PDT awareness, margin limits, FINRA-compliant logic)
- Trade logging & alerts (email notifications, timestamped records)
- Backtest report (full historical performance across 3+ years)
- Live demo testing (2-week forward test on demo before you go live)
- Revisions & support (we refine parameters based on your live results)
Claude builds #1. Professional builders deliver 1–8. The difference isn't "better coders." It's experience with what breaks in production. We've delivered 660+ projects on MQL5. We know what kills trading bots. We build around it by default.
Price: $100 for a simple breakout EA. $300+ for ICT/SMC-based strategies with multi-pair logic. AI/ML trading bots start at $350. Custom software from $500.
Best Case vs. Worst Case vs. What Actually Happens
Best case (professional): You hire a developer. They deliver a tested, documented EA. You deploy on a $5k account with 0.1 lot size. Over 6 months, you make $1,200 in live profit. The EA paid for itself. You scale to a $25k account. Profitable. Systematic. Sleeping better.
Worst case (Claude DIY): You build with Claude. It backtests perfectly. You deploy on a $3k account. Week 1: slippage and requotes hit differently than predicted. Day 7: you're down $1,100 (37% of your account). You panic. You turn it off. You waste another year wondering what went wrong. You never automate again.
The difference isn't luck. It's the invisible layer—risk management, broker knowledge, compliance, production testing—that separates toy code from professional systems.
Here's what happens next: You tell us your strategy. We show you a working demo in 45 minutes. We deploy live for 2 weeks on a demo account (real broker, real conditions). Then we go live on your account with micro risk. You give us 30 days of live data. We refine. Done.
Cost: $300–$500 upfront. Revisions included. Full transparency. Or build with Claude and hope the profit shows up.