Rank 114 · pullback → two-sided breakout window verdict · minimal clean replication

生成时间：2026-03-20 08:59 UTC

最小 clean replication 口径：固定 BTC/ETH/SOL 120d 15m 本地 cache；只挂 fib_retest_long 这 1 条 archetype；训练段冻结参数，测试段统一 signal 当根及之前数据 + next-bar open + no-overlap + hold 8 bars。

当前 hard verdict：park / evidence pool

测试段 6bps 下，冻结后的窗口版相对 baseline 的 mean_total_return 改善只有 -0.23%，retention 约 66.67%；更像靠砍样本换平滑，不足以继续占默认 fast lane。

冻结下来的唯一方案

• chosen_variant: window_pb3_w4_o20
• pullback_bars: 3
• window_bars: 4
• offset_atr: 0.2

测试段总表

variant	cost_bps_per_side	mean_total_return	positive_asset_ratio	mean_retention	mean_false_follow_4bars	mean_timeout_or_fail_rate	mean_entries	mean_expectancy
baseline_direct_entry	6.0	-0.72%	0.00%	100.00%	80.00%	0.00%	5.00	-0.14%
window_pb3_w4_o20	6.0	-0.95%	66.67%	66.67%	83.33%	33.33%	3.33	-0.52%
baseline_direct_entry	10.0	-1.12%	0.00%	100.00%	80.00%	0.00%	5.00	-0.22%
window_pb3_w4_o20	10.0	-1.21%	33.33%	66.67%	83.33%	33.33%	3.33	-0.60%
window_pb3_w4_o20	15.0	-1.55%	0.00%	66.67%	83.33%	33.33%	3.33	-0.70%
baseline_direct_entry	15.0	-1.61%	0.00%	100.00%	80.00%	0.00%	5.00	-0.32%

测试段分资产

asset	variant	cost_bps_per_side	signals_total	entries	retention	mean_total_return	post_cost_expectancy	false_follow_4bars_rate	timeout_or_fail_rate	win_rate	mean_best_move	mean_mae
BTC-USD	baseline_direct_entry	6.0	5	5	100.00%	-0.16%	-0.03%	100.00%	0.00%	40.00%	0.97%	-0.48%
BTC-USD	baseline_direct_entry	10.0	5	5	100.00%	-0.56%	-0.11%	100.00%	0.00%	40.00%	0.97%	-0.48%
BTC-USD	baseline_direct_entry	15.0	5	5	100.00%	-1.06%	-0.21%	100.00%	0.00%	40.00%	0.97%	-0.48%
BTC-USD	window_pb3_w4_o20	6.0	5	4	80.00%	0.47%	0.12%	75.00%	20.00%	50.00%	1.03%	-0.74%
BTC-USD	window_pb3_w4_o20	10.0	5	4	80.00%	0.15%	0.04%	75.00%	20.00%	50.00%	1.03%	-0.74%
BTC-USD	window_pb3_w4_o20	15.0	5	4	80.00%	-0.25%	-0.06%	75.00%	20.00%	50.00%	1.03%	-0.74%
ETH-USD	baseline_direct_entry	6.0	5	5	100.00%	-0.72%	-0.14%	80.00%	0.00%	40.00%	1.12%	-0.56%
ETH-USD	baseline_direct_entry	10.0	5	5	100.00%	-1.11%	-0.22%	80.00%	0.00%	40.00%	1.12%	-0.56%
ETH-USD	baseline_direct_entry	15.0	5	5	100.00%	-1.61%	-0.32%	80.00%	0.00%	40.00%	1.12%	-0.56%
ETH-USD	window_pb3_w4_o20	6.0	5	4	80.00%	0.09%	0.03%	75.00%	20.00%	25.00%	1.54%	-0.90%
ETH-USD	window_pb3_w4_o20	10.0	5	4	80.00%	-0.23%	-0.05%	75.00%	20.00%	25.00%	1.54%	-0.90%
ETH-USD	window_pb3_w4_o20	15.0	5	4	80.00%	-0.63%	-0.15%	75.00%	20.00%	25.00%	1.54%	-0.90%
SOL-USD	baseline_direct_entry	6.0	5	5	100.00%	-1.29%	-0.26%	60.00%	0.00%	20.00%	1.12%	-0.97%
SOL-USD	baseline_direct_entry	10.0	5	5	100.00%	-1.69%	-0.34%	60.00%	0.00%	20.00%	1.12%	-0.97%
SOL-USD	baseline_direct_entry	15.0	5	5	100.00%	-2.18%	-0.44%	60.00%	0.00%	20.00%	1.12%	-0.97%
SOL-USD	window_pb3_w4_o20	6.0	5	2	40.00%	-3.41%	-1.71%	100.00%	60.00%	0.00%	0.40%	-2.67%
SOL-USD	window_pb3_w4_o20	10.0	5	2	40.00%	-3.56%	-1.79%	100.00%	60.00%	0.00%	0.40%	-2.67%
SOL-USD	window_pb3_w4_o20	15.0	5	2	40.00%	-3.75%	-1.89%	100.00%	60.00%	0.00%	0.40%	-2.67%

训练段参数冻结网格（6bps）

暂无数据。

诚实边界

• 只用 signal 当根及之前已知的 OHLC / ATR / rolling fib 位置来定义 fib_retest_long。
• pullback bar 只允许在 signal 后前 1~3 根里寻找，且必须是当时已发生的 counter bar；不能回看更远 future 重选。
• success / failure / timeout 三态是一级结果；测试段统一 next-bar open 执行，禁止同 bar 判定 + 同 bar 成交。
• 本轮只回答 entry skeleton 值不值得留在 P1/P2 队列，不把它偷渡成独立 alpha。