How I Built My Crypto Trading Signal System with Claude

I’ve been trading crypto for two years. For most of that time I was doing everything manually — scanning charts, reading news, gut-feel entries. I wanted systematic signals but assumed I’d need a quant background or expensive tools to get there.

Then I started experimenting with Claude.

The problem I was trying to solve

Manual trading has one fatal flaw: you can’t watch the market 24/7. Crypto doesn’t close. I kept missing setups overnight or making emotional decisions after big moves.

What I wanted:

A system that monitors conditions I define
Fires a signal when conditions are met
Explains why it fired, not just that it fired

Step 1 — Describing my logic to Claude

I described my trading logic in plain language. No jargon — just exactly what I wanted:

“I want to buy BTC when: the 4-hour RSI crosses above 35 AND weekly volume is above the 20-period average AND the funding rate on Binance is negative.”

Claude turned this into working Python in about 4 minutes.

def check_signal(symbol='BTC/USDT'):
    df['rsi'] = RSIIndicator(df['close'], window=14).rsi()
    rsi_cross = df['rsi'].iloc[-2] < 35 and df['rsi'].iloc[-1] >= 35
    vol_ok = df['volume'].iloc[-1] > df['volume'].rolling(20).mean().iloc[-1]
    return rsi_cross and vol_ok

Step 2 — Using Extended Thinking to find edge cases

I asked Claude to find every scenario where the strategy would fire incorrectly. It identified three issues:

Bear trap on low-liquidity hours — RSI can cross 35 during Asian session with a single large order
Funding rate lag — Binance funding rates update every 8 hours, data can be stale
Consecutive signals — strategy could fire multiple times in one 4-hour window

Each of these would have cost me real money.

Step 3 — Backtesting

Results over 24 months of BTC 4-hour data:

47 signals fired
66% win rate
Average gain: +4.2%
Sharpe ratio: 0.87

What I’d do differently

Start with the edge cases, not the happy path. Ask for failure modes first.

Keep the strategy explainable. Simple rules beat optimized black boxes.