生成时间：2026-03-20 01:49 UTC ｜ 样本：BTC/ETH/SOL 120d 15m cache ｜ 口径：strict non-leaky early-window（先看前 3 根 post-break bars，再评估后 5 根管理价值）
硬结论：park / evidence pool
strict non-leaky early-window 版本只留下很弱的风险管理信号：low-flip 提前退出在 6bps/side 下仅带来约 0.16bps 的均值改善，主要价值更像轻微削尾而不是能稳定抬升 shared gate expectancy。
下一步：切 body-defined zone re-entry honest failure verdict 的 source intake；只有 fresh source 也 exhausted，才轮到 MTF CHOP > prebreak ladder > 旧 evidence pool。
close 突破 prev_high_20 / prev_low_20，且 body_ratio>=0.4、extension>=0.2 ATR。baseline_hold8 与 low_flip 提前在第 4 根 open 退出 的差别。| early_flip_bucket | events | share | mean_gross_ret_hold8 | mean_gross_ret_residual_5bars | cont_hit_rate_after_decision | fail_back_inside_rate_after_decision | tail_loss_p05_after_decision |
|---|---|---|---|---|---|---|---|
| high_flip_2 | 612 | 29.03% | 2.99 bps | 3.96 bps | 69.28% | 65.03% | -175.04 bps |
| low_flip_0 | 446 | 21.16% | 1.89 bps | -0.84 bps | 73.54% | 65.02% | -202.29 bps |
| mid_flip_1 | 1050 | 49.81% | 5.60 bps | -1.19 bps | 72.76% | 58.38% | -179.31 bps |
| side | early_flip_bucket | events | mean_gross_ret_hold8 | mean_gross_ret_residual_5bars | cont_hit_rate_after_decision | fail_back_inside_rate_after_decision |
|---|---|---|---|---|---|---|
| long | high_flip_2 | 284 | -8.57 bps | -3.58 bps | 67.61% | 66.20% |
| long | low_flip_0 | 222 | -0.32 bps | 0.93 bps | 69.37% | 63.96% |
| long | mid_flip_1 | 496 | 3.44 bps | 1.85 bps | 72.38% | 57.86% |
| short | high_flip_2 | 328 | 13.00 bps | 10.49 bps | 70.73% | 64.02% |
| short | low_flip_0 | 224 | 4.08 bps | -2.60 bps | 77.68% | 66.07% |
| short | mid_flip_1 | 554 | 7.53 bps | -3.91 bps | 73.10% | 58.84% |
| variant | cost_bps_per_side | events | mean_net_ret | tail_loss_p05 | positive_rate |
|---|---|---|---|---|---|
| baseline_hold8 | 6 | 2108 | -7.94 bps | -179.13 bps | 38.71% |
| baseline_hold8__long | 6 | 1002 | -12.79 bps | -181.11 bps | 37.33% |
| baseline_hold8__short | 6 | 1106 | -3.55 bps | -177.12 bps | 39.96% |
| overlay_exit_low_after_3bars | 6 | 2108 | -7.79 bps | -167.15 bps | 38.76% |
| overlay_exit_low_after_3bars__long | 6 | 1002 | -13.08 bps | -170.73 bps | 37.13% |
| overlay_exit_low_after_3bars__short | 6 | 1106 | -2.99 bps | -161.51 bps | 40.24% |
读法：若 low-flip 真的能稳定识别“脆弱延续”，那么 overlay_exit_low_after_3bars 应该明显优于 baseline。实际只看到极小的均值改善与轻微削尾，不足以升到 shared queue-facing gate。
reports/artifacts/scout_rank104_post_break_signflip_density_15m/event_log.csvreports/artifacts/scout_rank104_post_break_signflip_density_15m/bucket_summary.csvreports/artifacts/scout_rank104_post_break_signflip_density_15m/side_summary.csvreports/artifacts/scout_rank104_post_break_signflip_density_15m/policy_compare.csvreports/artifacts/scout_rank104_post_break_signflip_density_15m/verdict_summary.csv