Feature Iteration Run #577: zz_sylvain

# AUTOLOOP: zz_sylvain (3 iterations)

> **Phase**: Y4.7 Option B (DSL → Python code → FeatureSet candidate)
> **Target indicator**: zz_sylvain
> **Base feature_set_id**: 7
> **Claim**: (default) Propose a new k_* feature that improves risk-adjusted returns
> **Objective mode**: scalarize
> **Metrics**: ['profit_factor', 'sharpe_ratio', 'win_rate', 'max_drawdown', 'test_accuracy']
> **Total duration**: 87.43s
> **Total accepted**: 1
> **Total rejected**: 2

## Iterations 詳情

### Iteration 1
- DSL: `i_rsi(i_ema(k_close, 12) - i_ema(k_close, 26), 14)` (feature=k_loop_iter_1_128a6c)
- DSL valid: True
- FeatureSet candidate id: 684
- baseline PF: 1.286 → modified PF: 1.360
- t_stat: 0.000, p_value: 1.000, cohens_d: 0.000
- scalar_score: 0.4853
- accepted: True, reason: accepted

**Metrics summary:**
  - profit_factor: baseline=1.286, modified=1.360, diff=0.074
  - sharpe_ratio: baseline=7.879, modified=9.711, diff=1.832
  - win_rate: baseline=0.553, modified=0.574, diff=0.021
  - max_drawdown: baseline=0.075, modified=0.080, diff=0.005
  - test_accuracy: baseline=0.986, modified=0.986, diff=-0.000

### Iteration 2
- DSL: `(k_high.pct_change(12) - i_ema(k_high.pct_change(12), 9)) * k_volume.rolling(10).mean() / k_close` (feature=k_iter_2_k_high)
- DSL valid: False
- FeatureSet candidate id: None
- baseline PF: 0.000 → modified PF: 0.000
- t_stat: 0.000, p_value: 1.000, cohens_d: 0.000
- scalar_score: 0.0000
- accepted: False, reason: safe_eval DSL produces inf when ['k_close'] is zero; avoid unprotected division by these variables

### Iteration 3
- DSL: `k_high.pct_change(12) * dist_to_low - k_low.pct_change(12) * dist_to_high` (feature=k_loop_iter_3_663aed)
- DSL valid: True
- FeatureSet candidate id: 685
- baseline PF: 1.286 → modified PF: 1.286
- t_stat: 0.000, p_value: 1.000, cohens_d: 0.000
- scalar_score: 0.0000
- accepted: False, reason: p=1.000 ≥ 0.05, cohens_d=0.000 ≤ 0.3

**Metrics summary:**
  - profit_factor: baseline=1.286, modified=1.286, diff=0.000
  - sharpe_ratio: baseline=7.879, modified=7.879, diff=0.000
  - win_rate: baseline=0.553, modified=0.553, diff=0.000
  - max_drawdown: baseline=0.075, modified=0.075, diff=0.000
  - test_accuracy: baseline=0.986, modified=0.986, diff=0.000