当前硬结论：park
生成时间：2026-03-21 00:54 UTC | 口径：BTC/ETH/SOL, 15m, next-bar open, hold 8 bars, no-overlap, costs 6/10/15bps
最优 decoupled tolerance（train 60% 选定）= 0.8%。
本轮已完成最小 clean replication；与 baseline 相比，decoupled tolerance 在 6bps 只剩小幅改善，但 improvement 主要来自 long pocket，不是 desk 级 shared gate。
主弱点：uplift 主要集中在局部长侧 pocket；breakout_short 持续拖后腿，且跨资产仍分裂。
| tolerance_pct | trades | retention_vs_baseline | mean_return_bps |
|---|---|---|---|
| 0.300 | 41.00 | 0.37 | -22.96 |
| 0.500 | 69.00 | 0.62 | -13.06 |
| 0.800 | 86.00 | 0.77 | -6.30 |
| variant | cost_bps | trades | retention_vs_baseline | mean_return_bps | median_return_bps | win_rate | selected_tolerance_pct | return_delta_vs_baseline_bps |
|---|---|---|---|---|---|---|---|---|
| baseline | 6.00 | 75.00 | 100.00% | -1.86 | -21.69 | 37.33% | - | - |
| decoupled_tolerance_gate | 6.00 | 56.00 | 74.67% | -0.26 | -21.47 | 37.50% | 0.80 | 1.60 |
| baseline | 10.00 | 75.00 | 100.00% | -9.86 | -29.68 | 33.33% | - | - |
| decoupled_tolerance_gate | 10.00 | 56.00 | 74.67% | -8.26 | -29.46 | 32.14% | 0.80 | 1.60 |
| baseline | 15.00 | 75.00 | 100.00% | -19.86 | -39.65 | 29.33% | - | - |
| decoupled_tolerance_gate | 15.00 | 56.00 | 74.67% | -18.26 | -39.44 | 26.79% | 0.80 | 1.60 |
| asset | variant | cost_bps | trades | retention_vs_baseline | mean_return_bps | win_rate | selected_tolerance_pct | return_delta_vs_baseline_bps |
|---|---|---|---|---|---|---|---|---|
| BTC-USD | baseline | 6.00 | 29.00 | 100.00% | -10.62 | 41.38% | - | 0.00 |
| BTC-USD | decoupled_tolerance_gate | 6.00 | 24.00 | 82.76% | -14.77 | 41.67% | 0.80 | -4.16 |
| ETH-USD | baseline | 6.00 | 20.00 | 100.00% | -30.87 | 40.00% | - | 0.00 |
| ETH-USD | decoupled_tolerance_gate | 6.00 | 14.00 | 70.00% | 0.32 | 50.00% | 0.80 | 31.18 |
| SOL-USD | baseline | 6.00 | 26.00 | 100.00% | 30.21 | 30.77% | - | 0.00 |
| SOL-USD | decoupled_tolerance_gate | 6.00 | 18.00 | 69.23% | 18.65 | 22.22% | 0.80 | -11.57 |
| setup | variant | cost_bps | trades | retention_vs_baseline | mean_return_bps | win_rate | selected_tolerance_pct | return_delta_vs_baseline_bps |
|---|---|---|---|---|---|---|---|---|
| breakout_short | baseline | 6.00 | 25.00 | 100.00% | -48.76 | 20.00% | - | 0.00 |
| breakout_short | decoupled_tolerance_gate | 6.00 | 16.00 | 64.00% | -59.70 | 12.50% | 0.80 | -10.94 |
| ema_psar_long | baseline | 6.00 | 42.00 | 100.00% | 29.90 | 50.00% | - | 0.00 |
| ema_psar_long | decoupled_tolerance_gate | 6.00 | 32.00 | 76.19% | 34.91 | 53.12% | 0.80 | 5.01 |
| fib_retest_long | baseline | 6.00 | 8.00 | 100.00% | -22.04 | 25.00% | - | 0.00 |
| fib_retest_long | decoupled_tolerance_gate | 6.00 | 8.00 | 100.00% | -22.04 | 25.00% | 0.80 | 0.00 |
| metric | score_0_to_3 |
|---|---|
| usefulness | 1.00 |
| time_stability | 1.00 |
| cross_asset_stability | 1.00 |
| cost_trade_stability | 0.00 |
| deployability | 1.00 |
| rule_unclear | 0.00 |
| leakage_risk | 0.00 |
| post_cost_collapse | 1.00 |
| too_sparse | 0.00 |
| single_pocket_dependency | 1.00 |
| recommended_action | park |
| why_now | 本轮已完成最小 clean replication；与 baseline 相比，decoupled tolerance 在 6bps 只剩小幅改善，但 improvement 主要来自 long pocket，不是 desk 级 shared gate。 |
| main_weakness | uplift 主要集中在局部长侧 pocket；breakout_short 持续拖后腿，且跨资产仍分裂。 |