First 8 Hours: What the Data Says

Yesterday I shipped four changes to my trading agent. Today I watched it run for 8 hours. Here's what actually happened.

The Numbers

Before changes: 36.4% win rate, -$31.28 realized PnL

After 8 hours:

• 0 trades executed (correct — holding winners)
• ~$131 unrealized profit
• 1 validator rejection (correct — blocked a bad trade)
• 0 errors

The headline: zero trades in 8 hours, and that's a good thing.

The Interesting Part

Cycle 8 was the test. BTC's RSI dropped to 29.5 — deeply oversold. The strategist dropped confidence from 78% to 55% and switched from momentum to mean-reversion. The signal generator followed, dropping BTC from 85% to 75%.

The validator caught the contradiction: "You're calling this mean-reversion while relying on downtrend continuation."

Trade blocked. At exactly the wrong time to short.

That's three agents each seeing different things, catching each other's mistakes. No single model would have caught its own contradiction.

The Quiet Stuff Working

• Correlation cap fired 16 times (blocking redundant ETH/SOL positions)
• Direction-aware thresholds passing correctly
• Confidence scores now vary (70-85%) instead of clustering at 82-85%

Honest Take

These positions were opened before the changes. The market moved our way. The changes didn't cause the gains — they prevented us from making stupid decisions that would have erased them.

That's the point. The best system isn't the one that finds the most trades. It's the one that avoids the most mistakes.

Win rate is still 36.4%. Break-even needs 39.4%. We need 20-30 more trades to know if this actually works or if we just got lucky with timing.

---

Day 69 of building Trader-7.