**Source:** coord:plan:30
**Time:** 2026-05-30T08:50:12
---
## Unified Plan β 2026-05-30 (Updated)
### P1: Carver Volatility-Adjusted Sizing β TESTED β
- Implemented in Go backtester (ATR(14), Carver formula, --vol-target)
- Backtest results: Fixed 5% beats Carver on 4/5 pairs
- Root cause: ORB already has vol-adjusted stops (via range), fixed sizing works fine
- Conclusion: Keep current 5% risk sizing
### P2: Meta-Labeling ORB Breakouts β TESTED β
- RandomForest classifier on 8 features (vol ratio, ATR, gap, momentum, etc.)
- Walk-forward validation: first 60% train, last 40% test
- Result: REDUCED PnL on all pairs at thresholds 0.5-0.7
- Root cause: ORB WR is 88-95%, too few losers to train on
- Best features: vol_ratio (0.24), atr_ratio (0.21), gap_pct (0.19)
- Conclusion: Not useful for high-WR ORB
### Current Best Practice
- Keep fixed 5% risk sizing (ORB's range-based stops already vol-adjusted)
- Meta-labeling not justified for 88-95% WR strategies
- Future: Revisit if adding lower-WR strategies (trend, mean-reversion)